A respiratory detection system, method, electronic device and storage medium

Abstract

This application relates to the field of respiratory detection technology, and more particularly to a respiratory detection system, method, electronic device, and storage medium. The system includes: a PPG sensor for acquiring raw PPG data; a wavelet transform module for decomposing the raw PPG data into signal-related frequency band data to extract respiratory-related frequency band data; and a respiratory rate prediction module for feature extraction and respiratory rate prediction of the respiratory-related frequency band information to obtain target detection results. The wavelet transform module can effectively extract respiratory-related frequency band data from the PPG signal. This decomposition can distinguish physiological fluctuations of different frequencies caused by heartbeat and respiration, especially accurately capturing low-frequency changes caused by respiration, thereby improving the accuracy of respiratory rate prediction and ensuring the precision of target detection results.

Description

Technical Field

[0001] This application relates to the field of respiratory detection technology, and in particular to a respiratory detection system, method, electronic device and storage medium. Background Technology

[0002] Breathing monitoring is a crucial component of physiological surveillance, widely used in intensive care, sleep monitoring, and real-time health tracking. Traditional volumetric or pressure-based methods are used in clinical settings, including handheld spirometers, masks with flow sensors, and piezoelectric pressure-based breathing sensors. Additionally, breathing chest straps or bands are used to detect movement induced by breathing. While these methods offer high accuracy, they do not support simultaneous monitoring of other vital signs. Furthermore, their bulky and inconvenient design makes them unsuitable for long-term use, such as continuous monitoring during sleep or physical activity.

[0003] To detect respiratory signals non-invasively while ensuring patient comfort, photoplethysmography (PPG) has been widely proposed as an alternative method for respiratory monitoring. PPG is a non-invasive optical technique that measures changes in blood volume within blood vessels. This method involves irradiating the skin, typically using near-infrared or red light-emitting diodes. The sensor detects changes in light intensity caused by blood pulsation through the blood vessels and extracts the respiratory signal from the modulation of the pulse. PPG provides a portable, continuous respiratory monitoring solution while also monitoring other physiological signals such as heart rate and blood oxygen saturation, blood glucose levels, and arterial stiffness.

[0004] PPG (Positive Gas Perception) is used to detect respiration through several methods. The simplest approach is to directly measure respiration-related modulation by analyzing changes in signal peaks. However, this method is highly sensitive to noise and motion artifacts, which can significantly reduce its effectiveness in real-world scenarios. Frequency-based methods, such as Empirical Mode Decomposition (EMD) and holographic-Hilbert spectroscopy, are used to isolate specific respiratory signals from PPG signals, but they still require significant computational resources for hardware implementation. Furthermore, frequency-based methods cannot extract deeper, more complex features from the signal, limiting their ability to capture subtle changes caused by factors such as skin tone, age, and gender, which can lead to inaccuracies in PPG vital sign detection. Summary of the Invention

[0005] To address the inaccuracy of PPG vital sign detection mentioned above, this application provides a respiratory detection system, method, electronic device, and storage medium that can be applied to complex application scenarios and provides accurate detection.

[0006] A first aspect of this application provides a respiratory detection system, the system comprising:

[0007] PPG sensor, used to acquire raw PPG data;

[0008] The wavelet transform module is used to decompose the original PPG data into signals and extract frequency band data related to respiration.

[0009] The respiratory rate prediction module is used to extract features and predict the respiratory rate from the respiratory-related frequency band information to obtain the target detection result.

[0010] In some embodiments of this application, the PPG sensor includes a light source assembly, a photodetector, a signal amplifier, and an analog-to-digital converter.

[0011] In some embodiments of this application, obtaining raw PPG data includes:

[0012] Light is emitted through the light source assembly, and the reflected signal of the emitted light is received through the photodetector;

[0013] The reflected signal is amplified by the signal amplifier.

[0014] The amplified reflected signal is converted from analog to digital by the analog-to-digital converter to obtain the original PPG data.

[0015] In some embodiments of this application, the step of performing signal decomposition on the raw PPG data to extract respiration-related frequency band data includes:

[0016] The original PPG data is divided into even-indexed samples and odd-indexed samples;

[0017] Signal decomposition is performed based on the even-indexed and odd-indexed samples to extract frequency band data related to respiration.

[0018] In some embodiments of this application, the step of performing signal decomposition based on the even-numbered index samples and odd-numbered index samples to extract respiration-related frequency band data includes:

[0019] The first step is to calculate the difference between odd-indexed samples and even-indexed samples to obtain the detail coefficients;

[0020] The second step is to calculate the average of the odd-indexed samples and the even-indexed samples to obtain the approximation coefficient;

[0021] The third step involves dividing the approximation coefficients into even-indexed samples and odd-indexed samples, and then repeating the first and second steps in sequence until the frequency band data related to respiration is extracted.

[0022] In some embodiments of this application, the respiratory rate prediction module includes a convolutional layer, a pooling layer, and a fully connected layer.

[0023] In some embodiments of this application, the step of extracting features and predicting respiratory rate from the respiration-related frequency band information to obtain target detection results includes:

[0024] The convolutional layer is used to extract features from the respiration-related frequency band information to obtain a feature map;

[0025] The feature map is pooled using a pooling layer;

[0026] The results after pooling are integrated and analyzed through the fully connected layer to obtain a predicted respiratory rate, which is then used as the target detection result.

[0027] A second aspect of this application provides a method for detecting respiration, the method comprising:

[0028] Obtain raw PPG data;

[0029] The raw PPG data is decomposed to extract respiratory-related frequency band data;

[0030] Feature extraction and respiratory rate prediction are performed on the respiration-related frequency band information to obtain target detection results.

[0031] A third aspect of this application provides an electronic device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when executed by the processor, the processor causes the processor to perform the breathing detection method described in the embodiment.

[0032] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the respiratory detection method as described in the embodiments.

[0033] The technical solutions provided in this application embodiment have at least the following technical effects or advantages:

[0034] The respiratory detection system described in various embodiments of this application includes: a PPG sensor for acquiring raw PPG data; a wavelet transform module for decomposing the raw PPG data into signal and extracting frequency band data related to respiration; and a respiratory rate prediction module for feature extraction and respiratory rate prediction of the respiratory-related frequency band information to obtain target detection results. This application achieves an efficient frequency-based method by applying a wavelet transform module, filtering out key information related to respiratory signals, significantly reducing data volume, and improving the accuracy of neural network training. The PPG sensor can efficiently acquire raw PPG data, while the wavelet transform module is a temporally localized and computationally efficient module used to efficiently process raw PPG data, obtaining data closely related to respiration. Its multi-resolution characteristics enable it to simultaneously capture frequency bands used to calculate blood oxygen content and heart rate. Compared to other computationally intensive frequency-based techniques, the wavelet transform module is more efficient for real-time signal processing on resource-constrained devices because it can be implemented in hardware without multipliers or dividers. Unlike traditional discrete wavelet transform, the wavelet transform module does not rely on convolution but uses a lifting scheme to decompose the signal into different frequency bands.

[0035] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0036] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0037] Figure 1 This is a schematic diagram of the structure of a respiratory detection system according to an exemplary embodiment of this application;

[0038] Figure 2 This is a schematic diagram of another respiratory detection system in an exemplary embodiment of this application;

[0039] Figure 3 This is a schematic diagram comparing the signals collected by a respiratory sensor and a PPG sensor in an exemplary embodiment of this application;

[0040] Figure 4 This is a schematic diagram of an experimental testing environment setup in an exemplary embodiment of this application;

[0041] Figure 5 This is a schematic diagram of an experimental result in an exemplary embodiment of this application;

[0042] Figure 6This is a schematic diagram of the steps of a respiratory detection method in an exemplary embodiment of this application;

[0043] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an exemplary embodiment of this application.

[0044] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0045] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the embodiments depicted herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0046] Breathing monitoring is a crucial part of physiological monitoring, widely used in intensive care, sleep monitoring, and real-time health tracking. Traditionally, breathing monitoring is achieved through volume- or pressure-based methods, including handheld spirometers, masks with flow sensors, and piezoelectric pressure respiratory sensors. Additionally, breathing chest straps or bands are used to detect respiratory movements. While these traditional methods are highly accurate, they cannot simultaneously monitor other vital signs. Furthermore, their bulky and inconvenient design makes them unsuitable for long-term use, such as continuous monitoring during sleep or physical activity. To address these issues, photoplethysmography (PPG) has been proposed as an alternative breathing monitoring method. PPG is a non-invasive optical technique that acquires data by measuring changes in blood volume within blood vessels. This technique typically uses near-infrared or red light-emitting diodes to illuminate the skin; the sensor then detects changes in light intensity caused by blood pulsation and extracts the breathing signal from these changes. PPG is not only portable and supports continuous breathing monitoring, but it can also simultaneously monitor other physiological signals, such as heart rate, blood oxygen saturation, and even blood glucose levels and arterial stiffness.

[0047] The simplest way to detect respiratory signals using pulse-jet spectroscopy (PPG) is to directly measure respiratory-related modulation by analyzing changes in signal peak values. However, this method is highly sensitive to noise and motion artifacts, which may reduce its effectiveness in practical applications. Frequency-based methods, such as empirical mode decomposition and holographic-Hilbert spectroscopy, can isolate specific respiratory signals from PPG signals; however, these methods require significant computational resources for hardware implementation. Furthermore, frequency-based methods have limitations in extracting deep and complex features of the signal, making it difficult to accurately capture subtle changes due to skin tone, age, and gender, which may affect the accuracy of PPG vital sign detection.

[0048] To address the aforementioned technical problems, this application provides a respiratory detection system in its embodiments, such as... Figure 1 As shown, the system includes: a PPG sensor 101 for acquiring raw PPG data; a wavelet transform module 102 for decomposing the raw PPG data into a signal and extracting frequency band data related to respiration; and a respiratory rate prediction module 103 for extracting features and predicting the respiratory rate from the frequency band information related to respiration to obtain target detection results.

[0049] The PPG sensor plays a crucial role in this system, capturing physiological variables related to respiratory activity. In one specific implementation, the PPG sensor includes a light source assembly, a photodetector, a signal amplifier, and an analog-to-digital converter. Acquiring raw PPG data includes: emitting light through the light source assembly and receiving the reflected signal of the emitted light through the photodetector; amplifying the reflected signal through the signal amplifier; and converting the amplified reflected signal to digital data using the analog-to-digital converter to obtain the raw PPG data.

[0050] Specifically, such as Figure 2 As shown, LEDs are light source components. For example, red or near-infrared light generated by LEDs shines onto the user's skin. After the light is absorbed by the skin and blood vessels, some of the light is reflected back to the photodetector in the PPG sensor. Figure 2 The photodiode (PD) in the image is responsible for detecting the intensity of reflected light, which varies with the pulsation of blood flow. The reflected light signal is very weak, and is then amplified by a signal amplifier. Figure 2 The analog signal is amplified by a transimpedance amplifier (TIA). The amplified analog signal is then passed through an analog-to-digital converter (ADC). Figure 2 The analog voltage signal is converted into a digital signal by an ADC (Analog-to-Digital Converter). This step converts the analog voltage signal into digital form, facilitating subsequent digital processing. The digital signal is then transmitted via Bluetooth or other wireless technologies to a computing device (such as a smartphone or medical monitoring device) for further processing. In this embodiment, the raw PPG data in digital form is transmitted to the wavelet transform module. Figure 2 LWT.

[0051] Figure 2 In the analog front end, digitally controlled LEDs emit light onto the skin, and photodiodes (PDs) detect the reflected signals. The raw signal is processed by a transimpedance amplifier (TIA), converted into a digital signal by an analog-to-digital converter (ADC), and then transmitted to a digital device via Bluetooth. Figure 2The lower section illustrates the data processing flow: LWT decomposes the signal into frequency bands, and the LWT results can be further fed into a convolutional neural network (CNN) to extract features and classify and identify respiratory events (e.g., normal breathing, breath-holding, and asphyxia events). The CNN is then used to further process the signal to determine the respiratory rate and detect apnea. Furthermore, LWT can simultaneously isolate the frequency bands of SpO2 and heart rate (HR), making this system an ideal candidate for low-power, multifunctional wearable biosensor integration.

[0052] In one possible implementation, the original PPG data is decomposed to extract respiration-related frequency band data, including: dividing the original PPG data into even-indexed samples and odd-indexed samples; and performing signal decomposition based on the even-indexed and odd-indexed samples to extract respiration-related frequency band data. It is evident that the wavelet transform module uses an advanced algorithm to classify the original PPG data according to even and odd indices and processes them separately. This method effectively stratifies the variation characteristics of the time-series signal according to frequency, allowing for more targeted extraction of respiration-related frequency bands.

[0053] Specifically, the step of performing signal decomposition based on the even-indexed and odd-indexed samples to extract respiration-related frequency band data includes:

[0054] The first step is to calculate the difference between odd-indexed and even-indexed samples to obtain the detail coefficients. This step captures high-frequency components, including some noise and rapidly changing signal components, which is a particularly effective method for analyzing short-term dynamic changes.

[0055] The second step is to calculate the average of the even-indexed and odd-indexed samples to obtain an approximation coefficient. This coefficient represents the low-frequency components of the signal and is crucial for capturing longer-term physiological changes, such as the respiratory cycle.

[0056] The third step involves further decomposing the approximation coefficients into even and odd indices, repeating the previous detailed and approximate calculations layer by layer until a sufficient decomposition depth is reached, thereby extracting the frequency band data related to respiration. This multi-level decomposition can progressively refine the PPG data, extracting specific frequency band features related to different respiratory states.

[0057] It can be understood that the wavelet transform module, or LWT, performs frequency band decomposition on the original PPG signal to accurately extract frequency band data related to respiration. This module first divides the original PPG data into even-indexed samples and odd-indexed samples. Then, by calculating the differences and means between the samples, two types of coefficients are obtained: detail coefficients and approximation coefficients. Detail coefficients represent higher-frequency changes, mainly capturing rapid physiological changes, such as fluctuations caused by heartbeat; while approximation coefficients represent low-frequency changes, such as slow blood volume changes caused by respiration. Therefore, by using LWT to process the PPG signal, high-frequency noise is removed while achieving rapid and low-power extraction of blood oxygen, heart rate, and respiratory signals.

[0058] This decomposition process is iterative. In each iteration, the system further decomposes the approximate coefficients, gradually refining them to lower frequency bands until precise data reflecting respiratory-related frequency bands are obtained. This series of decompositions ultimately constitutes a set of frequency bands from high to low resolution, each reflecting changes in physiological activity at different time scales. Furthermore, the multi-resolution analysis properties of wavelet transform are particularly suitable for processing physiological signals with non-stationary characteristics, such as PPG signals. This method not only reduces data redundancy but also preserves key signal features, facilitating subsequent feature extraction and analysis.

[0059] The wavelet transform module's processing flow includes signal decomposition, step-by-step processing, and frequency band selection. During signal decomposition, such as... Figure 2 As shown, the wavelet transform module (LWT) is used to perform multi-level decomposition on the original PPG data. Each level of decomposition divides the signal into different frequency bands: a low-frequency band (approximation coefficients) and a high-frequency band (detail coefficients). The approximation coefficients at each level can be further decomposed to extract more detailed low-frequency components, while the high-frequency components contain detailed information related to heartbeat and respiration. The frequency range of each decomposition level in LWT can be calculated based on the sampling frequency Fs. The Nyquist frequency is F. Nyquist =Fs / 2, the detail coefficients in the frequency range of level L are:

[0060]

[0061] Here, "High" represents high frequency and "Low" represents low frequency. Finally, specific frequency bands (such as specific levels of detail coefficients) are selected. These frequency bands are considered to be closely related to the breathing pattern because they contain signals of periodic blood volume changes caused by respiration.

[0062] Figure 3 This is a schematic diagram comparing the signals collected by a respiratory sensor and a PPG sensor in an exemplary embodiment of this application. Figure 3 The image shows the actual respiratory signals collected from the respiratory sensor electrodes (top) and the raw data collected from the NIR channel of the PPG sensor (bottom), corresponding to... Figure 3 The data includes piezoelectric chip data and PPG infrared channel data. The carrier wave in the signal is generated by blood pulsation and is usually the more dominant signal, while the envelope wave is modulated by the blood pulsation signal caused by respiration.

[0063] In another specific implementation, the respiratory rate prediction module includes convolutional layers, pooling layers, and fully connected layers. Feature extraction and respiratory rate prediction are performed on the respiratory-related frequency band information to obtain target detection results. This includes: extracting features from the respiratory-related frequency band information through the convolutional layer to obtain a feature map; specifically, the convolutional layer processes the frequency band data after wavelet transform, extracts key time-frequency features through multiple filters, and generates a feature map. The pooling layer performs a pooling operation on the feature map, reducing its spatial size, which helps mitigate overfitting and improve the model's generalization ability. The fully connected layer further integrates the pooled data, using an activation function to increase non-linear processing capabilities, and finally outputs the predicted respiratory rate. In other words, the results after the pooling operation are integrated and analyzed through the fully connected layer to obtain a predicted respiratory rate value, which is used as the target detection result. The target detection result can be normal breathing, breath-holding, or asphyxiation events, etc.

[0064] Understandably, the respiratory rate prediction module further processes the frequency band data obtained through wavelet transform and uses a convolutional neural network (CNN) for deep learning analysis. First, the convolutional layers of this module extract features from the frequency band information using multiple filters, generating feature maps that centrally display key time-frequency features that may contain respiratory frequency information. Next, pooling layers downsample the feature maps, reducing data dimensionality and computational cost, while enhancing the model's insensitivity to small changes in the input data, helping to improve prediction accuracy and prevent overfitting. Finally, fully connected layers integrate all features and add non-linearity through activation functions such as ReLU to output the final predicted respiratory rate.

[0065] Overall, the design and implementation of this system enables accurate detection of respiratory rate, featuring low power consumption, high precision, and good user portability. For long-term monitoring, such as sleep quality analysis and daily health management, this system provides a highly practical solution.

[0066] To verify the technical effectiveness of the respiratory detection system described in this application, a series of experimental tests were also conducted. Figure 4 This is a schematic diagram of an experimental testing environment setup. Figure 4The test environment configuration is shown, including a dual-channel PPG sensor (red) located on the forehead and a reference sensor (blue) placed at the nostrils, with ground electrodes fixed to the chin. Six participants (four men and two women) were tested in the experiment. Each participant underwent four trials. In each trial, participants were asked to breathe normally for 30 seconds, hold their breath for 30 seconds, and then breathe again for one minute while data was collected. The experiment was conducted by wearing a PPG sensor and a piezoelectric pressure baseline respiratory sensor on the forehead, which senses changes in air pressure at the nostrils during breathing. Data from the PPG sensor was collected at 232 Hz, while data from the respiratory sensor was collected at 1000 Hz. To enhance comfort and reduce the impact of movement, both the PPG sensor and the respiratory sensor incorporated Bluetooth modules, and a soft bandage was used to securely adhere the sensors to the skin.

[0067] A total of 24 trials were collected, with each participant performing an average of 4 trials. Each trial collected PPG data from four channels: two red and two near-infrared (NIR) channels, and one data channel from a pressure sensor (left and right nostrils). Results were collected and analyzed using LWT. The PPG sensor used for respiratory detection consisted of four channels, utilizing two sets of LEDs, each including one red (660nm) and one NIR (940nm) LED. A piezoelectric pressure sensor was placed below the nose as a reference respiratory sensor. It measures respiratory rate by detecting small pressure changes caused by airflow during inhalation and exhalation. The experiment used reference data for neural network training and result validation.

[0068] Figure 5 This is a schematic diagram of an experimental result in an exemplary embodiment of this application, such as... Figure 5 As shown, the results from the PPG sensor and the respiratory sensor reveal a significant drop in respiratory signal power between 25 and 65 seconds, the period during which participants were asked to hold their breath. The power detected by the PPG sensor does not perfectly align with the readings of the dedicated respiratory sensor. This discrepancy occurs because the respiratory signal in PPG data is ultimately demodulated from a mixed signal, where blood vessel signals dominate. Furthermore, PPG data is more sensitive to noise due to its complex biological characteristics. Therefore, deep learning methods are needed to better extract information from this data.

[0069] Therefore, 18 trials were randomly selected for training, with each participant selecting 3 trials to increase variability. Each trial contained 4 datasets from four PPG sensor channels, each consisting of a 120-second test at 232Hz (approximately 27,840 data points). To further enhance the potential of the training data, jitter, amplitude modulation, and harmonic modulation were introduced to augment the data. Each participant retained six datasets for validation.

[0070] To demonstrate the system's ability to monitor respiration, its performance was evaluated using three metrics. First, respiratory rate per minute (BRM), defined as the number of breaths per minute, is the gold standard for comparison with advanced wearable respiratory sensors. Second, for the specific application of sleep monitoring and apnea detection, successful breath-holding detection measures the effectiveness of the algorithm in identifying breath-holding within a 30- to 60-second window. Breath-holding is defined as the signal amplitude dropping below a threshold set at half the mean of the absolute signal value. A duration of 20 seconds or more is considered 100% successful; shorter durations are assigned a success percentage relative to a 20-second detection time. The third metric is false apnea detection, used to evaluate instances of false breath-holding outside the window. A false detection occurs when the signal remains below the threshold for more than 5 seconds consecutively. Each instance outside the specified window is counted as a false detection, indicating that a low frequency of activity is misinterpreted as breath-holding. These conditions allow for the independent identification of genuine breath-holding events and false detections, regardless of the reference or real signal. As the threshold decreases, condition 1 decreases (undesirable), and condition 2 decreases (ideal), and vice versa.

[0071] For example Figure 5 As shown, in this experiment, the signal was collected at a frequency of 232 Hz. Therefore, the theoretically optimal range for calculating respiration is levels 8 and 9 of the detail coefficients, corresponding to a frequency range of 0.45 Hz to 0.91 Hz for level 8 and 0.23 Hz to 0.45 Hz for level 9. However, in practice, lower frequencies are more susceptible to motion artifacts, while higher frequencies are dominated by blood vessel signals modulated by the respiratory signal and may contain more useful information. By examining the waveforms and comparing the cross-correlation between each level (individual and combined layer summations) and the reference signal, it was found that the summation of detail coefficients from levels 5 to 7 was the most robust in our setup. Since each level of LWT reduces the total number of samples by half, the lower levels were upsampled to align with level 5. Compared to the original PPG data, the summation output was reduced to 1 / 32 of the original size while retaining the key low-frequency information needed to analyze respiratory patterns.

[0072] In addition to extracting low-frequency respiratory signals from high-level information, we also extract heart rate and blood oxygenation information from the middle layers (layers 3-6). This range overlaps somewhat with the range for respiratory detection, mainly to allow for some headroom, as subsequent calculation steps can handle a small amount of noise. The amplitude of the heart rate signal is also the AC signal amplitude of the PPG, which, by definition, is the change in reflected light intensity caused by pulse changes. The amplitude of the previously extracted respiratory signal is also the DC signal of the PPG, which is reflected light from skeletal muscle, skin, etc., which are not caused by pulses and are also modulated by low-frequency signals like respiration. After the heart rate frequency band is extracted, we mainly use moving averages and peak-valley calculations to confirm the frequency of this signal, thereby determining the heart rate. In addition, we use the formula R = (ACr / DCr) / (ACir / DCir) to calculate the R value and calculate arterial blood oxygen SpO2 by fitting. Here, ACr is the AC signal of red light, DCr is the DC signal of red light, and ACir and DCir are the AC and DC signals of infrared light, respectively.

[0073] As can be seen, by integrating PPG sensing technology with wavelet transform and deep learning analysis, the respiratory detection system provided in this application can continuously monitor respiratory status without contact with the user's body, reducing user discomfort and improving the continuity of data acquisition. In particular, wavelet transform not only reduces data redundancy but also preserves key signal features, which is helpful for subsequent feature extraction and analysis, thereby ensuring the accuracy of subsequent PPG vital sign detection. The low-power design and wireless data transmission capability of this system make it widely applicable, especially suitable for the medical and health fields that require long-term monitoring.

[0074] In some embodiments of this application, a respiratory detection method is also provided, such as... Figure 6 As shown, the method includes: S1, acquiring raw PPG data; S2, performing signal decomposition on the raw PPG data to extract respiration-related frequency band data; S3, performing feature extraction and respiratory rate prediction on the respiration-related frequency band information to obtain target detection results. The respiration detection method is a hybrid method that enhances the measurement of respiratory power through a wearable PPG device, integrating wavelet transform (LWT) and a three-layer convolutional neural network (CNN).

[0075] Step S2, or LWT, provides a computationally efficient, multi-resolution, and hardware-compatible method for processing raw PPG data while significantly reducing the data load on neural networks. Training the CNN enhances the system's ability to produce robust results, ultimately improving the accuracy of respiratory power measurement. Step S2 includes: dividing the raw PPG data into even-indexed and odd-indexed samples; and performing signal decomposition based on the even-indexed and odd-indexed samples to extract respiratory-related frequency band data. The signal decomposition based on the even-indexed and odd-indexed samples to extract respiratory-related frequency band data includes: a first step, calculating the difference between the odd-indexed and even-indexed samples to obtain detail coefficients; a second step, calculating the average of the odd-indexed and even-indexed samples to obtain approximation coefficients; and a third step, dividing the approximation coefficients into even-indexed and odd-indexed samples, repeating steps one and two sequentially until respiratory-related frequency band data is extracted.

[0076] Through the above multi-level decomposition, respiratory-related frequency band data can be effectively extracted from the PPG signal. This decomposition can distinguish physiological fluctuations of different frequencies caused by heartbeat and respiration, and in particular, it can accurately capture low-frequency changes caused by respiration.

[0077] In the feature extraction and respiratory rate prediction stages, a three-layer convolutional neural network (CNN) was used. Through its deep learning capabilities, the CNN can learn complex time-frequency features from the LWT-processed data, which are key to measuring respiratory power. The CNN structure includes multiple convolutional and pooling layers, enabling it to automatically detect and prioritize the signal features most meaningful for respiratory rate prediction. This approach not only improves prediction accuracy but also enhances the model's generalization ability to data under different physiological and environmental conditions.

[0078] This method effectively combines sensor acquisition, signal processing, and machine learning prediction, not only improving the accuracy of respiratory power measurement but also greatly expanding the application scope of monitoring technology due to its high efficiency and adaptability. Users can monitor their own respiratory status and effectively manage their respiratory health without the need for professional medical personnel. Furthermore, through continuous monitoring, the system can help identify potential respiratory-related health problems, such as sleep apnea, at an early stage.

[0079] Please refer to the following. Figure 7 This illustrates a schematic diagram of an electronic device provided by some embodiments of this application. For example... Figure 7As shown, the electronic device 2 includes: a processor 200, a memory 201, a bus 202, and a communication interface 203. The processor 200, the communication interface 203, and the memory 201 are connected via the bus 202. The memory 201 stores a computer program that can run on the processor 200. When the processor 200 runs the computer program, it executes the respiratory detection method provided in any of the foregoing embodiments of this application. The method includes: acquiring raw PPG data; performing signal decomposition on the raw PPG data to extract respiratory-related frequency band data; performing feature extraction and respiratory rate prediction on the respiratory-related frequency band information to obtain a target detection result.

[0080] The memory 201 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 203 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0081] Bus 202 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs. After receiving an execution instruction, the processor 200 executes the program. The control method disclosed in any of the foregoing embodiments of this application can be applied to the processor 200, or implemented by the processor 200.

[0082] The processor 200 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 200 or by instructions in software form. The processor 200 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 201. The processor 200 reads the information in memory 201 and, in conjunction with its hardware, completes the steps of the respiratory detection method.

[0083] This application also provides a computer-readable storage medium corresponding to the breathing detection method provided in the foregoing embodiments, wherein a computer program is stored thereon, and the computer program, when run by a processor, executes the breathing detection method provided in any of the foregoing embodiments.

[0084] In addition, examples of the computer-readable storage medium may include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be described in detail here.

[0085] In addition, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the respiration detection method provided in any of the foregoing embodiments. The method includes: acquiring raw PPG data; performing signal decomposition on the raw PPG data to extract respiration-related frequency band data; performing feature extraction and respiration rate prediction on the respiration-related frequency band information to obtain a target detection result.

[0086] Those skilled in the art will understand that the various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art should understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the virtual machine creation apparatus according to embodiments of this application.

[0087] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A respiratory detection system, characterized in that, The system includes: PPG sensor, used to acquire raw PPG data; The wavelet transform module is used to decompose the original PPG data into signals and extract frequency band data related to respiration. The respiratory rate prediction module is used to extract features and predict the respiratory rate from the respiratory-related frequency band information to obtain the target detection result; The step of performing signal decomposition on the raw PPG data and extracting respiration-related frequency band data includes: The original PPG data is divided into even-indexed samples and odd-indexed samples; Signal decomposition is performed based on the even-indexed and odd-indexed samples to extract frequency band data related to respiration.

2. The respiratory detection system according to claim 1, characterized in that, The PPG sensor includes a light source assembly, a photodetector, a signal amplifier, and an analog-to-digital converter.

3. The respiratory detection system according to claim 2, characterized in that, The acquisition of raw PPG data includes: Light is emitted through the light source assembly, and the reflected signal of the emitted light is received through the photodetector; The reflected signal is amplified by the signal amplifier. The amplified reflected signal is converted from analog to digital by the analog-to-digital converter to obtain the original PPG data.

4. The respiratory detection system according to claim 1, characterized in that, The signal decomposition based on the even-indexed and odd-indexed samples to extract respiration-related frequency band data includes: The first step is to calculate the difference between odd-indexed samples and even-indexed samples to obtain the detail coefficients; The second step is to calculate the average of the odd-indexed samples and the even-indexed samples to obtain the approximation coefficient; The third step involves dividing the approximation coefficients into even-indexed samples and odd-indexed samples, and then repeating the first and second steps in sequence until the frequency band data related to respiration is extracted.

5. The respiratory detection system according to claim 1, characterized in that, The respiratory rate prediction module includes convolutional layers, pooling layers, and fully connected layers.

6. The respiratory detection system according to claim 5, characterized in that, The step of extracting features and predicting respiratory rate from the respiration-related frequency band information to obtain target detection results includes: The convolutional layer is used to extract features from the respiration-related frequency band information to obtain a feature map; The feature map is pooled using a pooling layer; The results after pooling are integrated and analyzed through the fully connected layer to obtain a predicted respiratory rate, which is then used as the target detection result.

7. A method for detecting respiration, characterized in that, The method includes: Obtain raw PPG data; The raw PPG data is decomposed to extract respiratory-related frequency band data; Feature extraction and respiratory rate prediction are performed on the respiration-related frequency band information to obtain target detection results; The step of performing signal decomposition on the raw PPG data and extracting respiration-related frequency band data includes: The original PPG data is divided into even-indexed samples and odd-indexed samples; Signal decomposition is performed based on the even-indexed and odd-indexed samples to extract frequency band data related to respiration.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores computer-readable instructions, which, when executed by the processor, cause the processor to perform the respiratory detection method as described in claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the respiratory detection method as described in claim 7.

Images