A wide-range heart rate detection method based on piezoelectric thin film sensor

Abstract

The application provides a wide-range heart rate detection method based on a piezoelectric film sensor, and belongs to the technical field of physiological signal monitoring. The method comprises the following steps: signal acquisition and human existence determination; signal preprocessing; sliding window segmentation processing; short-time energy analysis; energy analysis based on step length multi-model; peak detection and data quality optimization; heart rate estimation; and multi-model decision fusion. The application significantly widens the heart rate detection range, has high sensitivity, strong anti-interference capability, is suitable for long-term non-sensing monitoring, and the like, and can be widely applied to intelligent mattresses, sleep monitoring, family health care and rehabilitation scenes.

Description

Technical Field

[0001] This invention relates to physiological signal monitoring technology, and in particular to a wide-range heart rate detection method based on a piezoelectric thin film sensor, applicable to smart mattresses, sleep monitoring systems and home health management devices. Background Technology

[0002] Heart rate refers to the frequency of the heartbeat. High heart rates are often closely related to physiological stress, anxiety, fever, arrhythmias, and even underlying cardiovascular diseases. Traditional heart rate monitoring methods, such as electrocardiograms (ECG), are highly accurate, but require electrodes and are complex to operate, making long-term, non-contact monitoring difficult. In recent years, non-contact heart rate detection technology based on cardiac impulse mapping (BCG) has received widespread attention due to its comfort and convenience. BCG signals are obtained by sensing the minute mechanical vibrations caused by the heartbeat and are often collected using piezoelectric thin-film sensors.

[0003] Among existing heart rate detection methods based on BCG signals, short-time energy methods are widely used due to their simple structure. However, their inherent single-scale analysis characteristics limit their performance in various practical scenarios: under high heart rate conditions, energy peaks often broaden or overlap, making it difficult to clearly distinguish adjacent heartbeats; in situations with weak signals or significant body motion interference, fixed-scale energy analysis is prone to producing spurious peaks, leading to period estimation errors; and in regions of rapid heart rate change, traditional methods struggle to maintain sufficient temporal resolution and cannot stably extract beat-by-beat heartbeat information. Therefore, traditional short-time energy methods still have significant shortcomings in covering a wide heart rate range. Thus, a new method is urgently needed that can achieve wide-range, highly robust heart rate monitoring while maintaining its non-invasive advantages. Summary of the Invention

[0004] To address the above problems, this invention proposes a wide-range heart rate detection method based on a piezoelectric thin-film sensor, the technical solution of which is as follows: Step 1: Signal Acquisition and Human Presence Determination. A piezoelectric thin-film sensor embedded within the mattress structure monitors changes in human contact pressure. When a stable human presence is determined within the detection area, micro-vibration signals are acquired to obtain the raw signal. The stability of the human body on the bed is determined by statistically analyzing the ratio of the DC component to the low-frequency energy of the piezoelectric thin-film output signal. When a human body is determined to be stably present for a consecutive 'a' seconds, the micro-vibration signal is acquired at a sampling rate of 200Hz and converted from analog to digital to obtain the raw digital signal, where a ≥ 20.

[0005] Step 2: Signal preprocessing. The original signal is detrended and filtered to extract the cardiac mechanical vibration component and obtain the BCG signal. The filtering uses a 4–16Hz finite impulse response (FIR) filter with an order of not less than 200, and a forward-backward bidirectional filtering method is used to eliminate phase distortion.

[0006] Step 3: Sliding window segmentation. The BCG signal is segmented according to a 20s sliding time window and a 1s step size to establish time series segments for subsequent energy analysis.

[0007] Step 4: Short-time energy analysis. Within each time window, a short-time energy analysis sequence with a sub-window length of 0.3s is constructed, and the energy sequence is smoothed by moving average.

[0008] Step 5: Energy analysis based on multiple models with different sliding sampling step sizes. Three short-time energy models based on different sliding sampling step sizes are constructed to perform energy analysis on the BCG signal, adaptively covering the heart rate detection range of 40–150 bpm. The short-time energy models include: (1) Model A: After performing interpolation enhancement on the BCG signal, the short-time energy is calculated with a step size of 3 sampling points and used for the medium to high heart rate range; the interpolation enhancement is achieved by linear interpolation or cubic spline interpolation; (2) Model B: Short-time energy is calculated with a step size of 2 sampling points and used in the low to medium heart rate range to suppress spurious peak interference at low heart rates; (3) Model C: Short-time energy is calculated with a step size of 1 sampling point, which is used for high time resolution detection in high heart rate range and fine-grained waveform capture under rapid heartbeat conditions.

[0009] Step 6: Peak Detection and Data Quality Optimization. Perform peak detection on the energy sequence output by the model to screen candidate heartbeat cycles with peak spacing between 80 and 300 sampling points; calculate the mean and standard deviation of the peak spacing set, and remove peak spacings that exceed the range of mean plus or minus standard deviation as outliers.

[0010] Step 7: Heart Rate Estimation. Calculate the heart rate cycle based on the effective peak interval and convert it to obtain the estimated heart rate value for the current time window.

[0011] Step 8: Multi-model decision fusion. A multi-model decision fusion mechanism is used to fuse the outputs of the three models. The mechanism automatically selects the optimal model based on the heart rate estimate of model A: when the output of model A is greater than a preset threshold and corresponds to a high heart rate (>110 bpm) range, the result of model C is selected as the final heart rate; otherwise, the result of model B is selected as the final heart rate.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a wide-range heart rate detection method based on a piezoelectric thin-film sensor, which can simply and accurately process and extract human vital sign data collected by the piezoelectric thin-film sensor, and can expand the heart rate measurement range to 40–150 bpm. This process and related algorithms perform well in embedded devices. Attached Figure Description

[0013] Figure 1 This is a basic flowchart of the present invention; Figure 2 It is the raw signal collected by the piezoelectric thin film sensor; Figure 3 It is a BCG signal obtained after filtering; Figure 4 It is a compact waveform of the energy signal at extremely high heart rates; Figure 5 It is an energy signal waveform accompanied by spurious peaks at extremely low heart rates. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0015] In this embodiment of the invention, the piezoelectric thin-film sensor is first firmly embedded inside the structural layer of the smart mattress (e.g., located below the latex layer or sponge layer, in the core area of ​​the human torso projection). When a person lies on the mattress, the mechanical pressure and the slight vibrations on the body surface caused by the heartbeat will cause the piezoelectric thin film to generate a change in charge, which is converted into a voltage signal through a preamplifier circuit and a conditioning circuit.

[0016] Combination Figure 1 The specific execution path of the method of this invention is described in detail: In step 1, the system continuously monitors changes in human body contact pressure and performs high-precision determination of the body's on-bed status. Since piezoelectric sensors are prone to baseline drift, this invention does not use a simple absolute amplitude threshold, but instead uses real-time statistical analysis of the ratio of the DC component of the piezoelectric film output signal to low-frequency energy (mainly the energy component caused by weak human breathing or vibration within the 0.5–5Hz frequency band). If the bed contains inanimate objects such as suitcases or heavy objects, the signal only shows a change in the DC component, without continuous low-frequency micro-vibration energy. In this case, the ratio is in an abnormal range, and the system remains in standby mode. Only when the ratio is within the preset stable range of human body on-bed characteristics, and satisfies the condition of being stable for a seconds (a≥20), does the system confirm that the human body is stably on the bed, and collects the human body micro-vibration signal with high precision at a sampling rate of 200Hz, thereby obtaining the original signal in digital format, such as... Figure 2 As shown.

[0017] In step 2, the acquired raw signal is preprocessed. Since the raw signal contains a large number of slowly varying trend terms caused by the mattress's own damping and high-frequency electromagnetic glitches from the outside, it is first de-trended. Subsequently, since the core frequency of the human heart mechanical vibration component (BCG signal) is stably distributed between 4–16 Hz, this invention uses a 4–16 Hz finite impulse response (FIR) filter for bandpass filtering. To suppress low-frequency interference caused by breathing, the order of this FIR filter is limited to no less than 200. To prevent nonlinear phase distortion caused by conventional filtering, this invention employs a forward-backward bidirectional filtering algorithm: that is, the signal is first forward filtered, and then the result is time-reversed before inverse filtering is performed. Since the phase delays introduced by forward and inverse filtering are equal in magnitude and opposite in direction, they cancel each other out in the time domain, thus achieving a "zero phase distortion" filtering effect and accurately extracting the pure BCG signal representing the heart mechanical vibration component, such as... Figure 3 As shown.

[0018] In step 3, a sliding window is used to segment the BCG signal in real time. The length of the sliding time window is set to 20 seconds, and the sliding step size is 1 second. Setting the large window to 20 seconds ensures that even under extremely low heart rate conditions of 40 bpm, each processing window can contain at least 10 complete heartbeat cycles, thus providing a sufficient sample size for subsequent statistical optimization and outlier removal; while the 1-second refresh step size ensures real-time update capability at the second level.

[0019] In step 4, short-time energy analysis is performed within each 20-second time window. To transform the high-frequency fluctuating BCG waveform into a smooth energy envelope, a short-time energy analysis sequence with a sub-window length of 0.3 seconds is constructed. After calculating the sum of squares of energy within each sub-window, a moving average smoothing process is further applied to the short-time energy sequence, effectively eliminating residual minor noise spikes and highlighting the clear energy peak envelope representing the heartbeat.

[0020] In step 5, based on the characteristics of biological signals, the heart rate range covers 40–150 bpm. At extremely high heart rates (e.g., >110 bpm), the heartbeat cycle is extremely short, and the waveform is extremely compact, such as... Figure 4 As shown; however, at extremely low heart rates (e.g., <50 bpm), the heart rate cycle is significantly prolonged, and secondary spurious peaks such as obvious recruitment waves are generated within a single diastolic phase, such as... Figure 5 As shown. To resolve this contradiction, this invention dynamically configures different sliding sampling step sizes, constructing three parallel energy analysis models: (1) Model A: First, the effective sampling rate of the original BCG signal is artificially doubled (from 200Hz to 400Hz) by using linear interpolation or cubic spline interpolation algorithm, thereby significantly enhancing the feature resolution of the medium to high heart rate range; on this basis, the short-time energy is calculated with 3 sampling points as the step size. This model serves as the basic model for full-band prior detection and is mainly used to capture the medium to high heart rate range. (2) Model B: Short-time energy is calculated using two sampling points as the step size. In the low to medium heart rate range, appropriately increasing the calculation step size is equivalent to applying an implicit low-pass smoothing effect, which can perfectly eliminate and suppress secondary spurious peak interference caused by cardiac diastole in the low heart rate cycle, and prevent the algorithm from misjudging spurious peaks as heartbeats; (3) Model C: Short-time energy is calculated directly using a single sampling point as the extremely fine-grained step size. This model is specifically designed for the high heart rate range. Under rapid heartbeat conditions, the 1-point step size provides extremely high instantaneous temporal resolution, enabling fine-grained waveform capture and ensuring that compact heartbeat energy peaks are not missed.

[0021] In step 6, peak detection and data quality optimization are performed on the smoothed energy sequence output by the multi-model system. First, the system applies an adaptive amplitude threshold to perform a round of peak localization and filters out candidate heartbeat cycles with peak spacing between 80 and 300 sampling points. At a raw sampling rate of 200 Hz, the instantaneous heart rate corresponding to 80 sampling points is 150 bpm, and the instantaneous heart rate corresponding to 300 sampling points is 40 bpm. This design directly filters out extremely short spurious peak spacing caused by body movement or extremely long abnormal spacing caused by missed detection. Subsequently, to cope with single-point interference caused by sudden changes in respiratory depth, the system calculates the mean μ and standard deviation σ of all candidate peak spacing sets within the current 20-second time window. According to the mean-standard deviation statistical criterion, individual peak spacings exceeding the range [μ-σ, μ+σ] are directly identified as sudden outliers and removed.

[0022] In step 7, based on the optimized set of effective peak intervals, its precise mean is calculated, and the heart rate estimate for the current 20-second time window is obtained.

[0023] In step 8, multi-model decision fusion is performed. The system uses the heart rate estimate calculated by model A as a bridge for adaptive switching: a preset heart rate fusion switching threshold is set (preferably 110 bpm in this embodiment). When the preliminary heart rate estimate calculated by model A is greater than 110 bpm, it indicates that the human body is currently in a high heart rate or tachycardia state. At this time, the system decision chain automatically switches to and selects the output result of model C, which has extremely high temporal resolution, as the final heart rate. If the preliminary result of model A is not greater than 110 bpm, it indicates that the human body is currently in the normal resting or bradycardia range. At this time, the decision chain automatically switches to and selects the output result of model B, which has extremely strong pseudo-peak suppression capability, as the final heart rate. Through this fusion decision mechanism, high-precision adaptive detection within the full frequency band of 40–150 bpm is perfectly achieved.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A wide-range heart rate detection method based on a piezoelectric thin-film sensor, characterized in that, The specific steps are as follows: Step 1: Monitor changes in human contact pressure using a piezoelectric thin film sensor located inside the mattress structural layer. Collect micro-vibration signals of the human body when the human body is determined to be stably present in the detection area, and obtain the raw signal. Step 2: Perform detrending and filtering on the original signal to extract the cardiac mechanical vibration component and obtain the BCG signal; Step 3: Divide the BCG signal into segments according to the set sliding time window and step size to establish time series segments; Step 4: Construct a short-time energy analysis sequence for each time window and perform smoothing on the energy sequence; Step 5: Based on different heart rate zones, construct multiple short-time energy models with different sampling step sizes, and perform energy analysis on the BCG signal; Step 6: Perform peak detection on the energy sequence output by the model to obtain candidate heartbeat cycles, and use statistical criteria to remove abnormal peak spacing to suppress interference; Step 7: Calculate the heart rate cycle based on the effective peak interval and convert it to obtain the heart rate estimate for the current time window; Step 8: Use a multi-model decision mechanism to fuse the heart rate estimation results of various short-time energy models and output the final heart rate.

2. The method according to claim 1, characterized in that, In step 1, the stability of the human body in bed is determined by statistically analyzing the ratio of the DC component of the piezoelectric film output signal to the low-frequency energy. When the human body is determined to be stable within a consecutive a seconds, the micro-vibration signal is collected at a sampling rate of 200Hz, where a≥20.

3. The method according to claim 1, characterized in that, In step 2, a bandpass filter is used to extract the cardiac mechanical vibration component; the bandpass filter is a 4–16Hz FIR filter with an order of not less than 200, and a forward-backward bidirectional filter is used to eliminate phase distortion.

4. The method according to claim 1, characterized in that, In step 3, the sliding time window is set to 20s with a step size of 1s; in step 4, a short-time energy analysis sequence with a sub-window length of 0.3s is constructed and a moving average smoothing is performed.

5. The method according to claim 1, characterized in that, In step 5, the three short-time energy models constructed based on different sampling step sizes include: Model A: After interpolating and enhancing the BCG signal, short-time energy is calculated with a step size of 3 sampling points, which is used for the moderate to high heart rate range; the interpolation enhancement is achieved through linear interpolation or cubic spline interpolation; Model B: Short-time energy is calculated with a step size of 2 sampling points and is used in the low to medium heart rate range to suppress spurious peak interference at low heart rates; Model C: Calculates short-time energy with a step size of 1 sampling point, used for high temporal resolution detection in high heart rate zones and fine-grained waveform capture under rapid heartbeat conditions; The multi-model design is based on a dynamic configuration of the sliding step size, enabling the detection of heart rates to cover the range of 40–150 bpm.

6. The method according to claim 1, characterized in that, Step 6 specifically involves: performing peak detection on the energy sequence output by the model, screening candidate heartbeat cycles with peak spacing between 80 and 300 sampling points; calculating the mean and standard deviation of the peak spacing set, and removing peak spacings that exceed the range of mean plus or minus standard deviation as outliers.

7. The method according to claim 1, characterized in that, The multi-model decision-making mechanism in step 8 is as follows: the optimal model is automatically selected based on the heart rate estimate of model A. When the output of model A is greater than the preset threshold and corresponds to a high heart rate zone, the result of model C is selected as the final heart rate. Otherwise, select the result of model B as the final heart rate.

Images