An ECG heart rate measurement adaptive denoising method and system based on ambulance cabin environment sensor explicit reference

Abstract

This invention provides an adaptive denoising method and system for ECG heart rate measurement based on explicit reference from environmental sensors within the ambulance cabin. The method includes: introducing existing triaxial accelerometers and vibration sensors within the ambulance cabin as explicit noise reference sources into the ECG signal processing flow; optimizing the parameter adjustment logic of traditional Kalman filtering; optimizing the QRS wave detection logic; and standardizing the definition of all adjustable parameters. The system uses a high-performance vehicle-mounted edge computing unit as its core computing unit, meeting the requirements for wide-temperature operation in industrial-grade vehicles. This invention effectively separates motion artifact noise with the same frequency as the QRS wave through explicit environmental noise reference input and an adaptive filtering mechanism. Through baseline correction and environmental cross-validation, it retains the steep characteristics of the QRS wave while achieving strong denoising, balancing denoising capability and physiological waveform fidelity. The end-to-end processing latency is less than 20ms, meeting the real-time requirements of pre-hospital emergency care. No additional hardware modifications are required, and it can be batch adapted to various smart ambulance devices.

Description

Technical Field

[0001] This invention relates to the field of biomedical signal, edge computing and 5G medical IoT interdisciplinary technology, and in particular to an adaptive noise reduction method and system for ECG heart rate measurement based on explicit reference of environmental sensors in an ambulance cabin. Background Technology

[0002] 5G smart ambulances are the core carriers of the pre-hospital emergency care system. They can collect patients' vital signs data in real time through onboard intelligent devices, enabling early assessment of the patient's condition and emergency preparation, significantly shortening the golden rescue time. Currently, mainstream smart ambulances are equipped with vital sign monitors, three-axis accelerometers, cabin vibration sensors, 5G communication gateways, and edge computing devices as standard equipment, providing the hardware foundation for simultaneous acquisition and real-time processing of multi-source data.

[0003] However, road bumps, engine vibrations, and patient limb movements during ambulance travel generate a large amount of motion artifact noise. This type of noise is concentrated in the 5-20Hz frequency range, which highly overlaps with the core frequency of the human QRS wave (5-15Hz), constituting typical co-frequency interference. Existing vehicle-mounted ECG processing algorithms generally employ fixed-parameter filtering and traditional Pan-Tompkins detection algorithms, which have three prominent problems: First, they only process the ECG single signal and cannot distinguish between co-frequency noise and the true QRS wave. In high-bump scenarios, the false detection rate of QRS waves exceeds 50%, resulting in severe distortion of heart rate measurements. Second, there are problems with the Kalman filter parameter adjustment logic. Some solutions have insufficient adaptability to the operating conditions and cannot adapt to bumpy conditions. Third, the parameters are all fixed adjustment values ​​without standardized calibration basis, making them unsuitable for different hardware models and exhibiting extremely poor versatility.

[0004] Existing technologies do not fully utilize the environmental sensor resources on ambulances, do not establish a correlation mapping between noise and ECG signals, and lack a standardized parameter calibration system and reasonable filtering adjustment logic. They cannot meet the core requirements of strong noise reduction and preservation of physiological characteristics, which seriously restricts the accuracy and stability of pre-hospital heart rate monitoring. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an adaptive denoising method and system for ECG heart rate measurement based on explicit reference from environmental sensors inside the ambulance cabin. By using explicit environmental noise reference input and an adaptive filtering mechanism, motion artifact noise with the same frequency as the QRS wave is effectively separated. Through baseline correction and environmental cross-validation, the steep characteristics of the QRS wave are preserved while achieving strong denoising, balancing denoising capability and physiological waveform fidelity. The end-to-end processing latency is less than 20ms, meeting the real-time requirements of pre-hospital emergency care. No additional hardware modifications are required, and it can be batch adapted to various smart ambulance devices.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, comprising the following steps:

[0007] S1 Multi-source Data Synchronous Acquisition: Using a uniform sampling frequency f s The system simultaneously acquires single-lead ECG signals from patients, data from triaxial accelerometers within the ambulance cabin, and environmental noise data from vibration sensors. A unified sampling trigger pulse is generated using the general-purpose GPIO timer output pin of the onboard edge computing unit, and then connected in parallel to the external trigger input interfaces of each sensor via signal distribution cables to achieve microsecond-level time synchronization. When a sensor lacks an external trigger interface, a software timestamp alignment method is used, with the edge computing unit's high-precision hardware clock uniformly timestamping the arrival time of the data, achieving synchronization jitter ≤50μs. Data from each channel is then written to a circular buffer after being aligned according to the timestamp.

[0008] S2 Extended State Explicit Input Kalman Filtering for Noise Reduction: A three-dimensional extended state vector is constructed, containing the pure ECG signal, the first derivative of the ECG signal, and the vehicle vibration noise component. The synthesized triaxial acceleration value and the measured value of the vibration sensor are used as the external control input for filtering, and a linear mapping relationship between cabin vibration interference and ECG signal transmission is established. A linear Kalman filter algorithm with fixed parameter process noise covariance matrix and dynamic update of observation noise variance is adopted to recursively separate and extract artifact-free pure ECG signal point by point.

[0009] S3 Global Filter Parameter Adaptive Adjustment: Before updating the filter at each sampling point, the overall global environmental noise intensity index σ inside the cabin is extracted through a sliding window of M1 sampling points. k The variance of the basic observation noise is calculated based on the real-time noise intensity; the greater the intensity of the turbulence vibration inside the cabin, the larger the value of the basic observation noise variance, and the intensity of the basic filtering and noise reduction of the algorithm is improved accordingly.

[0010] S4 Environmental Coupled QRS Wave Fidelity Protection: Based on filtered prior estimates, a stable ECG baseline is extracted and the relative rate of change of signal mutations is calculated to screen candidate ECG mutation locations; instantaneous in-cabin vibration data from M3 sampling points are retrieved to conduct environmental cross-verification, distinguishing physiological QRS waves from vibration noise spikes; for segments determined to be true QRS waves, the observation noise variance of the current and subsequent sampling points is dynamically reduced in combination with the mutation confidence level to avoid smoothing and destroying QRS characteristic waves;

[0011] S5 Heart Rate Calculation: The QRS wave position is identified on the pure ECG signal after fidelity processing, and the average value of the RR interval of the continuous NRR group is calculated to obtain the real-time heart rate.

[0012] In a preferred embodiment, in claim S2, the three-dimensional extended state vector is defined as follows:

[0013] Propose a three-dimensional extended state vector x k :

[0014]

[0015] Among them, s k Let ṡ be the estimated value of the pure ECG signal at the k-th sampling time. k Let n be the first time-domain derivative of the ECG signal at the k-th sampling time. k Let s0 be the vehicle vibration noise state quantity coupled to the ECG channel at the k-th sampling time. Obviously, when k=0, s0=ṡ0=n0=0;

[0016] A two-dimensional external control input vector u is proposed. k :

[0017]

[0018] Among them, a k The magnitude of the combined triaxial acceleration vector, v k Real-time sampling values ​​for cabin vibration sensors;

[0019] Propose discrete linear state transition equations:

[0020]

[0021] Where Φ represents the time-domain state transition matrix and B represents the vibration noise control mapping matrix;

[0022] Propose a single-channel observation equation:

[0023]

[0024] Among them, z k For the single-channel observation of the filtering model, w corresponds to the original noisy ECG signal acquired by the monitor at the k-th sampling time. k For three-dimensional state processes, Gaussian white noise, e k This is residual Gaussian observation noise in the ECG hardware channel.

[0025] In a preferred embodiment, the filter coefficient matrix is ​​fixed as follows:

[0026] Propose a time-domain state transition matrix :

[0027]

[0028] Propose a vibration and noise control mapping matrix :

[0029]

[0030] Propose an ECG coupling observation matrix :

[0031]

[0032] Where Δt=1 / f s h is the system sampling time interval. a h is the acceleration-coupled ECG noise transmission coefficient. v The transmission coefficient of ECG noise coupled to cabin vibration.

[0033] In a preferred embodiment, in S2, the expression for the global fixed process noise covariance matrix Q is:

[0034]

[0035] Where q0 is the basic process noise figure of the electrocardiogram, q n The noise figure is a process noise figure specific to the vibration and noise state. Both parameters are obtained through offline calibration of hardware parameters.

[0036] In a preferred embodiment, the overall cabin noise intensity σ k Calculation formula:

[0037]

[0038] in, For the acceleration-coupled ECG noise transmission coefficient, The vibration-coupled ECG noise transmission coefficients are used to convert acceleration and vibration sensor readings into equivalent noise voltages for the ECG channel, achieving dimensional uniformity. α is the relative weighting coefficient for acceleration noise contribution, and β is the relative weighting coefficient for vibration noise contribution; both are dimensionless and calibrated through pre-hospital vehicle-based fitting under actual operating conditions, with a recommended value of α + β = 1. M1 is the sliding window length (number of sampling points), corresponding to a duration of 1 second, i.e., M1 = f s × 1s; aᵢ is the composite amplitude of the triaxial acceleration vector at the i-th sampling time (unit: m / s²); vᵢ is the real-time sampling value of the cabin vibration sensor at the i-th sampling time (unit: mV).

[0039] In a preferred embodiment, in S3, an adaptive dynamic observation noise scalar variance calculation formula is proposed:

[0040]

[0041] Where, r s,0 K represents the inherent noise floor variance of the vehicle-mounted ECG channel under static conditions. adapt σ is the global adaptive adjustment coefficient. maxThe maximum noise level is preset for extreme driving conditions of ambulances.

[0042] In a preferred embodiment, in S4, a screening process for candidate sites of electrocardiographic mutations is proposed:

[0043] For the ECG signal estimated by Kalman filtering prior, the ECG baseline b is extracted by sliding median filtering at M2 point. k Calculate the baseline removal signal s′ k =s k -b k ,s′ k To remove baseline drift from the ECG signal (i.e., the baseline-de-baseline signal), the unit is mV; the baseline mean amplitude μ is iteratively updated using an exponential moving average algorithm. b,k Iteration formula:

[0044]

[0045] Where, α ema α is the smoothing coefficient for the exponential moving average. ema = 1 / (f s ×10), corresponding to a 10-second baseline tracking time constant.

[0046] Further solve for the relative abrupt change rate of the signal:

[0047]

[0048] Preset fixed mutation threshold T r When the relative mutation rate r of the real-time signal k >T r At that time, the current sampling point is determined to be a candidate site for ECG mutation, which is used for subsequent environmental cross-verification.

[0049] In a preferred embodiment, the environment cross-validation method in S4 is as follows:

[0050] Calculate the instantaneous environmental stability index τ of M3 sampling points at the corresponding time of the candidate event. k :

[0051] like < If the candidate event is a true QRS wave, it is determined to be a turbulence noise spike; otherwise, it is determined to be a turbulence noise spike, and the normal filter strength is maintained. To ensure environmental stability and safety margin, take =3×√(α·h a ²·r a,0 +β·hᵥ²·r v,0 The physical meaning of ) is the upper limit of 3σ after the sensor's background noise is mapped to the ECG channel through the noise transfer coefficient.

[0052] In a preferred embodiment, in S4, the formulas for calculating the abrupt change confidence of the true QRS wave and the final filter parameters are as follows:

[0053]

[0054]

[0055] Among them, c k K represents the mutation confidence level, with a value range of [0,1]. prot R is the QRS wave protection strength coefficient; k final K represents the observation noise variance ultimately used for filter updates. prot The QRS protection strength coefficient is set between 0.80 and 0.95 (recommended value: 0.90; the closer the value is to 1, the stronger the QRS protection and the more complete the waveform is preserved, but excessive protection will cause noise to remain in the QRS segment).

[0056] This invention provides an adaptive denoising system for ECG heart rate measurement based on explicit reference of environmental sensors inside an ambulance cabin. The system executes the aforementioned adaptive denoising method for ECG heart rate measurement based on explicit reference of environmental sensors inside an ambulance cabin, and includes a multi-source data acquisition module and an on-board edge computing module that are electrically and communicatively connected in sequence.

[0057] The multi-source data acquisition module includes an on-board five-parameter vital signs monitor, a three-axis acceleration acquisition unit, and a cabin vibration acquisition unit. The hardware synchronously triggers the acquisition of three types of time-series homogeneous data: electrocardiogram, acceleration, and vibration.

[0058] The vehicle-mounted edge computing module adopts an industrial wide-voltage vehicle-mounted edge computing power all-in-one machine with a total AI computing power of 275 TOPS. It is equipped with 64GB LPDDR5 vehicle-specific memory and has built-in embedded algorithms for three-dimensional extended state Kalman filtering, noise adaptive control, QRS wave verification, and heart rate calculation, realizing full-process local low-latency signal processing.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. Significantly improved noise reduction performance: Correcting errors in traditional filtering logic, using in-cabin accelerometers and vibration sensors as explicit noise reference sources, establishing a linear mapping relationship between vibration interference and ECG artifacts, effectively filtering out motion artifacts with the same frequency as the QRS wave under high turbulence conditions, and improving noise residue and waveform distortion problems.

[0061] 2. Accurate and stable heart rate measurement: ECG baseline is extracted by sliding median filtering to eliminate the interference of QRS wave amplitude on threshold calculation. Cross-validation with instantaneous environmental vibration data is used to distinguish the real QRS wave from turbulence noise spikes. Under strong noise reduction conditions, QRS wave characteristics are fully preserved, meeting the requirements of pre-hospital emergency heart rate monitoring.

[0062] 3. Zero hardware modification cost: It relies entirely on the existing standard hardware of 5G smart ambulances and can be deployed simply by upgrading the algorithm software, making it suitable for mass promotion and application.

[0063] 4. Real-time performance meets emergency needs: The algorithm adopts a Kalman filter architecture that iteratively predicts each sampling point. It performs state prediction and update only once per sampling point, with a computational complexity of O(1) constant level. The end-to-end processing latency is lower than the 100ms real-time red line for pre-hospital emergency care, making it suitable for deployment on vehicle-mounted edge computing devices.

[0064] 5. High versatility: A complete parameter calibration specification is established, and all parameters are automatically derived based on the hardware manual or offline calibration method, making it compatible with different models of monitors, accelerometers, and vibration sensors. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of an ECG denoising and heart rate measurement framework based on an environmental sensor reference, according to a preferred embodiment of the present invention. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, 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 pertains.

[0068] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0069] An adaptive denoising method for ECG heart rate measurement based on explicit reference from environmental sensors inside the ambulance cabin. Figure 1 This includes the following steps:

[0070] Step 1: Synchronous Collection of Multi-Source Data

[0071] According to the sampling frequency specified in the equipment manual By utilizing the idle GPIO pins of the in-vehicle edge computing all-in-one machine, it is programmed and configured to output in hardware timer PWM mode to generate a periodic signal. A square wave pulse signal (50% duty cycle, 3.3V LVTTL level) is simultaneously connected to the SYNC pin of the triaxial accelerometer and the EXT_TRIG pin of the vibration sensor acquisition module, triggering both to start sampling at the same time with a deviation of no more than 1 μs.

[0072] The ECG module of the monitor samples autonomously using its own internal clock, and the data is sent frame by frame to the edge computing all-in-one machine via UART or USB interface.

[0073] After the data from the three channels arrives at the all-in-one machine, each channel is timestamped using the machine's own high-precision hardware clock: accelerometer and vibration sensor data are timestamped in the GPIO interrupt service routine, while ECG data is timestamped in the serial port receive completion interrupt. All three channels share the same hardware clock, allowing for direct timestamping comparison.

[0074] The filtering algorithm matches frames by timestamp: for each frame of acceleration data (timestamp) ) and vibration data (timestamp) ), select the frame with the closest timestamp from the ECG data sequence, if Then, the three frames form a set of aligned data. Since all three channels use the same sampling frequency... The work involves two equally spaced grids: the ECG timestamp sequence and the acceleration / vibration timestamp sequence (both with a spacing of [missing information]). For any given frame of acceleration data, a deviation of no more than [a certain value] can always be found. There are no alignment issues with the corresponding ECG frames. Channel alignment accuracy: no more than 1 μs between acceleration and vibration, and no more than 50 μs between ECG and both (depending on interrupt response jitter). Hz ( For a system with a time accuracy of ms, a deviation of 50 μs is less than 3% of one sampling period, which meets the engineering requirements for multi-channel time alignment in subsequent filtering.

[0075] Step 2: Explicit Noise Reference Kalman Filter Preprocessing

[0076] Before starting the filter, complete the following initialization settings:

[0077] (1) Initial state vector

[0078]

[0079] Physical meaning: At the moment of filter activation, the amplitude, rate of change, and vibration noise components of the pure ECG signal are all set to zero. This assumption is based on the AC coupling characteristics of the ECG signal—the monitor's front-end hardware has filtered out the DC bias, and the baseline is zero when there is no signal input.

[0080] (2) Initial covariance matrix

[0081]

[0082] The initial covariance matrix is ​​in diagonal form, with off-diagonal elements set to zero (assuming the initial estimation errors of each state component are independent). The assignment principles and recommended values ​​for the three diagonal elements are detailed in Section II, "Supplementary Table of Initialization Parameters." The core principle is that the initial uncertainty is relatively large—the values ​​of each element of P0 are slightly larger than the state variance during steady-state operation, making the filter more confident in the observations (z) during the startup phase. k This allows us to quickly approach the true state.

[0083] (3) Initial observation noise variance

[0084]

[0085] Using the inherent noise floor variance of the vehicle-mounted ECG channel under static conditions As the initial observation noise variance ( The calibration method is shown in Section II, Main Parameter Table. After the adaptive filter is started, R... k Starting from the first sampling point, the real-time ambient noise intensity σ is used as the basis. k Dynamically updated, R0 only applies to the first sampling point.

[0086] (4) Convergence rate analysis

[0087] With a relatively large P0 setting, the elements in the initial Kalman gain K0 have large values, and the posterior state estimation for the first sampling point is mainly driven by the observed values. Subsequent sampling points are then recursively calculated one step, with the covariance matrix P... k The predicted update cycle rapidly decays to a steady-state value. Consider f. s = 500 Hz typical sampling rate, after about 510 sampling points (1020 ms) P k The process of entering steady state and the convergence of state estimation is much shorter than the cycle of a single heartbeat (approximately 750 ms), and does not affect the real-time performance of heart rate calculation.

[0088] Step 3: Global Noise Adaptive Parameter Adjustment

[0089] After the state prediction of each sampling point is completed and before the filtering update, the overall vibration and noise intensity σ inside the cabin is calculated through a sliding window of M1 with a duration of 1 second.k The variance R of the baseline observation noise at the current moment is dynamically calculated based on the noise intensity. k When the intensity of the bumps in the ambulance increases, σ k When R increases, k The synchronous increase reduces the reliance on the original noisy ECG signal, thus strengthening the filtering and noise reduction efforts; when the cabin environment is stable, σ k Decrease, R k Synchronous reduction preserves effective ECG details, achieving adaptive equalization noise reduction.

[0090] Step 4: Baseline Correction and Environmentally Sensing QRS Wave Protection

[0091] The pure baseline of the ECG was extracted using a 200ms M2 sliding median filter to eliminate the interference of the QRS wave's own amplitude on the baseline mean, and the relative rate of change r of the signal was accurately calculated. k All candidate events for waveform abrupt changes were screened out. This is to prevent prolonged absence of QRS waves (such as cardiac arrest) from causing μ... b,k Continue to decay to zero, r k Abnormal amplification, for μ b,k Set lower limit protection: when μ b,k <μ min At that time, forced μ b,k =μ min (μ recommended) min =0.05 mV). μ under normal sinus rhythm b,k If the value is much higher than this threshold, the protection will not be triggered, and r will not be changed. k The structure of the calculation formula.

[0092] Instantaneous environmental stability indicators were calculated using a 100ms M3 window. Cross-validation was performed: the mutation under stable environmental conditions was determined to be a true QRS wave, and the confidence level c was calculated. k It also reduces the filtering intensity while fully preserving the steep characteristics of the QRS wave; abrupt changes under turbulent conditions are identified as noise spikes, maintaining a strong filtering and noise reduction state to effectively filter out artifacts.

[0093] Environmental cross-validation employs differential judgment logic: 100ms instantaneous environmental indicators are used. Compared with the 1s global noise baseline Compare - if < This indicates that no sudden turbulence or impact exceeding the operating baseline occurred within the current 100ms timeframe, and the candidate abrupt change can be determined as a true QRS wave; if ≥ This indicates that the instantaneous vibration intensity deviates significantly from the current environmental baseline, and the candidate abrupt change is most likely caused by a bump or impact. Therefore, the normal filtering intensity should be maintained. This design enables the QRS wave protection mechanism to function under various operating conditions, including stationary, smooth driving, and mild bumps, triggering protection suppression only when a sudden bump or impact overlaps with a signal abrupt change in time.

[0094] Step 5: Heart Rate Calculation

[0095] The QRS wave position was detected on the final optimized and denoised ECG signal. NRR intervals were collected and averaged, then substituted into the heart rate formula to calculate the real-time heart rate. The heart rate conversion used the industry standard formula:

[0096]

[0097] In the formula, HR represents real-time heart rate in bpm; RR i N represents the interval between adjacent valid QRS waves. RR The number of RR intervals participating in the average.

[0098] Hardware device description

[0099] This invention requires no additional hardware modification and is compatible with standard hardware in conventional 5G smart ambulances: a five-parameter vital signs monitor, a three-axis accelerometer, a cabin vibration sensor, a high-performance vehicle-mounted edge computing all-in-one machine, a 5G high-speed vehicle-mounted gateway, and a cloud-based private cloud platform.

[0100] The core specifications of the in-vehicle edge computing all-in-one machine include: a 12-core industrial-grade embedded general-purpose processor adapted to the wide temperature and voltage operating environment of vehicles; an integrated dedicated AI acceleration module containing over two thousand parallel computing cores and dozens of tensor operation acceleration units, achieving an overall AI computing power of 275 TOPS; 64GB LPDDR5 memory (102.4GB / s bandwidth) and 1TB NVMe SSD high-speed storage, stably supporting real-time signal processing and AI inference tasks; support for 9–36VDC wide voltage input; a fanless compact design; and compliance with industrial-grade wide temperature requirements of -10~50℃, serving as the core for local signal processing in vehicles.

[0101] All parameters in this invention are not fixed empirical values, but can be directly calculated and calibrated based on the nominal parameters specified in the corresponding hardware device manual. The specific rules are as follows:

[0102]

[0103]

[0104]

[0105]

[0106]

[0107] Noise Transmission Coefficient Calibration Instructions

[0108] Noise transmission coefficient h a h v The calibration was obtained using offline calibration fitting on an on-board vibration table. The calibration process is as follows:

[0109] The ECG acquisition module of the vital signs monitor was fixed to a standard vibration table fixture, and multiple sets of fixed-frequency standard vibration excitations with known amplitudes and frequencies covering the typical bumpy range of an ambulance were applied; the effective value of the triaxial acceleration A under each set of excitations was acquired simultaneously. i , Vibration sensor sampling effective value V i And the amplitude Δs of ECG motion artifacts induced under the corresponding operating conditions. i The acceleration-ECG noise transmission coefficient h was obtained by fitting the relationship between the vibration excitation amplitude and the ECG artifact amplitude through linear regression. a With vibration-ECG noise transmission coefficient h v The fitting equation is:

[0110]

[0111]

[0112] After calibration, the two coefficients are written as fixed parameters into the vibration and noise control mapping matrix and will not be changed during the filtering process.

[0113] h a h v The calibration accuracy directly affects the filtering algorithm's performance on s k With n k The separation quality. Observation equation z k = s k + n k +e k In the middle, s k With n k The single-channel observations are mathematically not uniquely decomposable, and the algorithm's separation capability relies entirely on environmental sensors through B·u. k For n k Independent prior information is provided. Calibration deviations will be proportionally transmitted to the ECG clean signal estimation deviation, therefore the relative calibration error should be controlled within ±15% (it is recommended that the vibration table regression R² ≥ 0.90 be used as a calibration quality reference).

[0114] Specific suggestions:

[0115] (1) When calibrating the vibration table, the excitation of each group should be repeated at least 3 times and the average value should be taken to reduce random error;

[0116] (2) After calibration, verify the ECG reconstruction residuals under static and known bumpy conditions;

[0117] (3) If vibration table calibration cannot be carried out due to limitations (such as in a rapid on-site deployment scenario), the following conservative value selection strategy can be adopted: h a Take values ​​of 0.15–0.25, h v Use values ​​of 0.15–0.25 (1.5–2.5 times the recommended value of 0.10, with both coefficients scaled up proportionally, maintaining the acceleration-dominant design of α=0.8 and β=0.2 from the original parameter table). Larger values ​​cause the filter to "over-rely on environmental sensors and enhance noise reduction" when precise calibration is lacking—the prior prediction amplitude of the environmental sensors is too large → the filter attributes a larger proportion of the energy in the ECG observation to vibration noise → resulting in stronger noise reduction. Although this may slightly weaken ECG details, it avoids false positives in the QRS complex due to residual noise. In pre-hospital emergency scenarios, it is better to have a slightly smoother signal than to misinterpret noise as a heartbeat. Iterate and refine the values ​​using a vibration table once conditions are suitable.

Claims

1. An adaptive noise reduction method for ECG heart rate measurement based on explicit reference from environmental sensors inside an ambulance cabin, characterized in that, Includes the following steps: S1 Multi-source Data Synchronous Acquisition: Using a uniform sampling frequency f s The system simultaneously acquires single-lead ECG signals from patients, data from triaxial accelerometers within the ambulance cabin, and environmental noise data from vibration sensors. A unified sampling trigger pulse is generated using the general-purpose GPIO timer output pin of the onboard edge computing unit, and then connected in parallel via signal distribution cables to the external trigger input interfaces of each sensor to achieve microsecond-level time synchronization. When a sensor lacks an external trigger interface, a software timestamp alignment method is used, with the edge computing unit's high-precision hardware clock uniformly timestamping the data arrival time, achieving synchronization jitter ≤50μs. Data from each channel is then written to a circular buffer after being aligned according to the timestamp. S2 Extended State Explicit Input Kalman Filtering for Noise Reduction: A three-dimensional extended state vector is constructed, containing the pure ECG signal, the first derivative of the ECG signal, and the vehicle vibration noise component. The synthesized triaxial acceleration value and the measured value of the vibration sensor are used as the external control input for filtering, and a linear mapping relationship between cabin vibration interference and ECG signal transmission is established. A linear Kalman filter algorithm with fixed parameter process noise covariance matrix and dynamic update of observation noise variance is adopted to recursively separate and extract artifact-free pure ECG signal point by point. S3 Global Filter Parameter Adaptive Adjustment: Before updating the filter at each sampling point, the overall global environmental noise intensity index σ inside the cabin is extracted through a sliding window of M1 sampling points. k The variance of the basic observation noise is calculated based on the real-time noise intensity; the greater the intensity of the turbulence vibration inside the cabin, the larger the value of the basic observation noise variance, and the intensity of the basic filtering and noise reduction of the algorithm is improved accordingly. S4 Environmental Coupled QRS Wave Fidelity Protection: Based on filtered prior estimates, a stable ECG baseline is extracted and the relative rate of change of signal mutations is calculated to screen candidate ECG mutation locations; instantaneous in-cabin vibration data from M3 sampling points are retrieved to conduct environmental cross-verification, distinguishing physiological QRS waves from vibration noise spikes; for segments determined to be true QRS waves, the observation noise variance of the current and subsequent sampling points is dynamically reduced in combination with the mutation confidence level to avoid smoothing and destroying QRS characteristic waves; S5 Heart Rate Calculation: The QRS wave position is identified on the pure ECG signal after fidelity processing, and the average value of the RR interval of the continuous NRR group is calculated to obtain the real-time heart rate.

2. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 1, is characterized in that... In right S2, the three-dimensional extended state vector is defined as follows: Propose a three-dimensional extended state vector x k : Among them, s k Let ṡ be the estimated value of the pure ECG signal at the k-th sampling time. k Let n be the first time-domain derivative of the ECG signal at the k-th sampling time. k Let s0 be the vehicle vibration noise state quantity coupled to the ECG channel at the k-th sampling time. Obviously, when k=0, s0=ṡ0=n0=0; A two-dimensional external control input vector u is proposed. k : Among them, a k The magnitude of the combined triaxial acceleration vector, v k Real-time sampling values ​​for cabin vibration sensors; Propose discrete linear state transition equations: Where Φ represents the time-domain state transition matrix and B represents the vibration noise control mapping matrix; Propose a single-channel observation equation: Among them, z k For the single-channel observation of the filtering model, w corresponds to the original noisy ECG signal acquired by the monitor at the k-th sampling time. k For three-dimensional state processes, Gaussian white noise, e k This is residual Gaussian observation noise in the ECG hardware channel.

3. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 2, is characterized in that... The filter coefficient matrix is ​​configured as follows: Propose a time-domain state transition matrix : Propose a vibration and noise control mapping matrix : Propose an ECG coupling observation matrix : Where Δt=1 / f s h is the system sampling time interval. a h is the acceleration-coupled ECG noise transmission coefficient. v The transmission coefficient of ECG noise coupled to cabin vibration.

4. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 1, is characterized in that... In S2, the expression for the global fixed process noise covariance matrix Q is: Where q0 is the basic process noise figure of the electrocardiogram, q n The noise figure is a process noise figure specific to the vibration and noise state. Both parameters are obtained through offline calibration of hardware parameters.

5. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 1, is characterized in that... Overall cabin noise intensity σ k Calculation formula: in, For the acceleration-coupled ECG noise transmission coefficient, The vibration-coupled ECG noise transmission coefficients are used to convert acceleration and vibration sensor readings into equivalent noise voltages for the ECG channel, achieving dimensional uniformity. α is the relative weighting coefficient for acceleration noise contribution, and β is the relative weighting coefficient for vibration noise contribution; both are dimensionless and calibrated through pre-hospital vehicle-based fitting, with a recommended value of α+β=1. M1 is the sliding window length, corresponding to a duration of 1 second, i.e., M1= f s × 1s; aᵢ is the composite amplitude of the three-axis acceleration vector at the i-th sampling time; vᵢ is the real-time sampling value of the cabin vibration sensor at the i-th sampling time.

6. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 5, is characterized in that... In S3, a formula for calculating the scalar variance of adaptive dynamic observation noise is proposed: Where, r s,0 K represents the inherent noise floor variance of the vehicle-mounted ECG channel under static conditions. adapt σ is the global adaptive adjustment coefficient. max The maximum noise level is preset for extreme driving conditions of ambulances.

7. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 1, is characterized in that... In S4, a screening process for candidate sites of electrocardiographic mutations is proposed: For the ECG signal estimated by Kalman filtering prior, the ECG baseline b is extracted by sliding median filtering at M2 point. k Calculate s′ k =s k -b k ,s′ k To remove baseline drift from the ECG signal, the baseline mean amplitude μ is iteratively updated using an exponential moving average algorithm. b,k Iteration formula: Where, α ema α is the smoothing coefficient for the exponential moving average. ema = 1 / (f s ×10), corresponding to a 10-second baseline tracking time constant; Further solve for the relative abrupt change rate of the signal: Preset fixed mutation threshold T r When the relative mutation rate r of the real-time signal k >T r At that time, the current sampling point is determined to be a candidate site for ECG mutation, which is used for subsequent environmental cross-verification.

8. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 7, is characterized in that... S4 environment cross-validation method: Calculate the instantaneous environmental stability index τ of M3 sampling points at the corresponding time of the candidate event. k : like < If the candidate event is a true QRS wave, it is determined to be a turbulence noise spike; otherwise, it is determined to be a turbulence noise spike, and the normal filter strength is maintained. To ensure environmental stability and safety margin, take =3×√(α·h a ²·r a,0 +β·hᵥ²·r v,0 The physical meaning of ) is the upper limit of 3σ after the sensor's background noise is mapped to the ECG channel through the noise transfer coefficient.

9. The adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside the ambulance cabin, as described in claim 8, is characterized in that... In S4, the formulas for calculating the abrupt change confidence of the true QRS wave and the final filter parameters are as follows: Among them, c k R represents the mutation confidence level, with values ​​ranging from [0,1]. k final K represents the observation noise variance ultimately used for filter updates. prot The protection strength coefficient for QRS waves ranges from 0.80 to 0.

95.

10. An adaptive noise reduction system for ECG heart rate measurement based on explicit reference from environmental sensors inside an ambulance cabin, characterized in that, An adaptive noise reduction method for ECG heart rate measurement based on explicit reference of environmental sensors inside an ambulance cabin, as described in any one of claims 1 to 9, includes a multi-source data acquisition module and an on-board edge computing module connected in sequence by electrical communication. The multi-source data acquisition module includes an on-board five-parameter vital signs monitor, a three-axis acceleration acquisition unit, and a cabin vibration acquisition unit. The hardware synchronously triggers the acquisition of three types of time-series homogeneous data: electrocardiogram, acceleration, and vibration. The vehicle-mounted edge computing module adopts an industrial wide-voltage vehicle-mounted edge computing power all-in-one machine with a total AI computing power of 275 TOPS. It is equipped with 64GB LPDDR5 vehicle-specific memory and has built-in embedded algorithms for three-dimensional extended state Kalman filtering, noise adaptive control, QRS wave verification, and heart rate calculation, realizing full-process local low-latency signal processing.

Images