A method and device for early warning of sleep apnea in newborns

Abstract

This application discloses a method and device for early warning of sleep apnea in newborns, relating to the field of intelligent monitoring and pattern recognition technology. The method includes: acquiring initial audio data and real-time monitoring audio data of a newborn's sleep; performing feature analysis on the initial audio data to construct a standard breathing pattern, i.e., a dominant frequency signal, characterizing individual breathing features; dividing the real-time monitoring audio data into multiple time segments, calculating the pattern matching degree between each time segment and the standard breathing pattern to identify variations in the breathing waveform; and classifying and discriminating based on the matching degree to generate a risk warning signal for sleep apnea. This application achieves accurate identification and early warning of newborns' sleep breathing state by constructing personalized standard breathing patterns and using pattern matching algorithms to quantify the differences between real-time data and the standard pattern.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring and pattern recognition technology, specifically to a method and device for early warning of sleep apnea in newborns. Background Technology

[0002] Newborns, especially premature infants and low birth weight infants, are prone to sleep apnea (defined as cessation of breathing lasting more than 20 seconds, or less than 20 seconds but accompanied by symptoms such as cyanosis and bradycardia) due to their immature respiratory systems. If not detected and intervened in time, it may lead to hypoxic-ischemic encephalopathy, intellectual disability, or even death. Currently, commonly used clinical early warning methods include polysomnography (PSG) and pulse oximetry (POS). While PSG can comprehensively monitor multi-dimensional signals such as electroencephalography (EEG), electrooculography (EOG), and respiratory airflow, it is bulky, complex to operate, and invasive, making it unsuitable for long-term continuous monitoring and home use. POS, on the other hand, relies on blood perfusion status, which can lead to data distortion in newborns with low perfusion, and it only provides feedback after hypoxia occurs, resulting in a delayed warning.

[0003] In existing technologies, monitoring technologies based on acoustic sensors (e.g., microphones) have the advantages of being portable, comfortable, and easy to operate, making them suitable for daily home monitoring scenarios. They utilize highly sensitive miniature microphones to collect sound frequency data such as the airflow sound and the faint sound of chest rise and fall during a newborn's breathing, and then identify whether the breathing rhythm is normal based on the sound frequency data. However, when monitoring with acoustic sensors, the breathing sound of a newborn is inherently weak and varies greatly from person to person, making it easily affected by environmental noise, such as the sound of household appliances running and family members talking. This results in a low accuracy rate for extracting and identifying breathing signals. At the same time, it is difficult to distinguish breathing signals from non-breathing acoustic signals such as limb friction and clothing shaking, making it difficult to stably capture low-amplitude, short-cycle breathing sounds, which can easily lead to false alarms or missed alarms. Summary of the Invention

[0004] The purpose of this application is to provide a method and device for early warning of sleep apnea in newborns, so as to solve the technical problem that existing methods for monitoring sleep apnea in newborns based on newborn audio data are prone to false alarms or missed alarms.

[0005] To achieve the above objectives, this application provides the following technical solution: Firstly, this application proposes a technical solution for a neonatal sleep apnea early warning method, which includes: Acquire initial and monitor audio data of the newborn during sleep; the initial audio data is acquired when the newborn is in a stable and normal breathing state. Based on the initial audio data, a dominant frequency signal is obtained; the dominant frequency signal is the signal generated by the breathing frequency of the newborn during sleep. Based on the monitored audio data, multiple time-series data segments are obtained; Based on the main frequency signal and each time series data segment, a matching degree corresponding to each time series data segment is obtained; the matching degree is at least used to characterize the degree of similarity between the main frequency signal and the corresponding time series data segment. Based on each matching degree, a sleep risk value is obtained; the sleep risk value is used to characterize at least the probability of the newborn experiencing apnea. If the sleep risk value is greater than or equal to the first preset value, an early warning will be issued.

[0006] As a specific solution in this application, the acquisition of initial audio data of newborn sleep includes: Acquire first audio data; the first audio data is audio data that meets first preset conditions; the first preset conditions include: the first audio data is audio data collected by a microphone from a preset area; the preset area includes an area 10 cm to 30 cm in front of or to the side of the newborn's mouth and nose; Based on the first audio data, second audio data is obtained; the second audio data is the audio data in the first audio data that meets the second preset condition; Based on the second audio data, the initial audio data is obtained.

[0007] As a specific solution in this application, the step of obtaining the main frequency signal based on the initial audio data includes: Based on the initial audio data, multiple first component signals are obtained; the first component signals are signals obtained from the initial audio data based on empirical mode decomposition. The target signal is obtained by traversing each first component signal; the target signal is any signal in each first component signal for which no corresponding reference degree has been obtained; the reference degree is used at least to characterize the probability that the target signal is the main frequency signal. Based on the target signal, obtain the reference degree corresponding to the target signal; After each first component signal has acquired its corresponding reference degree, the main frequency signal is acquired based on each first component signal; the main frequency signal is the signal with the highest corresponding reference degree among each first component signal.

[0008] As a specific solution in this application, obtaining the reference degree corresponding to the target signal includes: The target signal is fitted to obtain a fitting curve; Based on the fitted curve, a first period and a first variance are obtained; the first period is the minimum positive period of the fitted curve; the first variance is the sum or average of the variances of each maximum value and the variances of each minimum value in the fitted curve. Based on the first period, a first difference is obtained; the first difference is equal to the absolute value of the difference between the first period and the second period; the second period is equal to the average value of the minimum positive period of the fitted curves corresponding to each first component signal. Based on the first difference and the first variance, the reference degree corresponding to the target signal is obtained; the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance.

[0009] As a specific solution in this application, the step of obtaining the matching degree corresponding one-to-one with each time-series data segment based on the main frequency signal and each time-series data segment includes: Traverse each time series data segment to obtain the current data segment; the current data segment is any data segment in each time series data segment for which no corresponding matching degree has been obtained; Based on the current data segment and the main frequency signal, a first similarity is obtained; the first similarity is the similarity between the current data segment and the main frequency signal. Based on the first similarity, the matching degree corresponding to the current data segment is obtained.

[0010] As a specific solution in this application, the step of obtaining the first similarity based on the current data segment and the main frequency signal includes: Based on the current data segment and the main frequency signal, a first average value, a second average value, and a second similarity are obtained; the first average value is the average value of each extreme point after normalization of the current data segment; the second average value is the average value of each extreme point after normalization of the main frequency signal; the second similarity is negatively correlated with the dynamic time warping distance between the current data segment and the main frequency signal. A second difference is obtained based on the first average and the second average; the second difference is equal to the absolute value of the difference between the first average and the second average. The first similarity is obtained based on the second difference and the second similarity; the first similarity is negatively correlated with the second difference, and the first similarity is positively correlated with the second similarity.

[0011] As a specific solution in this application, obtaining the matching degree corresponding to the current data segment based on the first similarity includes: Based on the current data segment, multiple second component signals are obtained; the second component signals are signals obtained from the current data segment based on empirical mode decomposition. Multiple signal pairs are obtained based on multiple first component signals and multiple second component signals; each first component signal is a signal obtained from the initial audio data based on empirical mode decomposition; the absolute value of the Pearson correlation coefficient of the first component signal and the second component signal in the signal pair is greater than or equal to a third preset value; Based on each signal pair, a first quantity and a second quantity are obtained; the first quantity is the total quantity of each signal pair; the second quantity is the total quantity of each first component signal. Based on the first quantity and the second quantity, a quantity ratio is obtained; the quantity ratio is the ratio of the first quantity to the second quantity. Based on the first similarity and the quantity ratio, the matching degree corresponding to the current data segment is obtained.

[0012] As a specific solution in this application, the step of obtaining sleep risk values ​​based on various matching degrees includes: Based on each matching degree, a first matching degree and a preset number of second matching degrees are obtained; the timing of each second matching degree and the first matching degree is sequentially adjacent, and the first matching degree is the matching degree corresponding to the last data segment in each timing data segment; Fit each second matching degree to the first matching degree to obtain a fitted straight line; Based on the fitted line, the slope of the line is obtained; the slope of the line is the slope of the fitted line. The sleep risk value is obtained based on the slope of the straight line and the first matching degree; the first matching degree is the matching degree corresponding to the last data segment in each time series data segment.

[0013] As a specific solution in this application, after obtaining the sleep risk value based on the slope of the straight line and the first matching degree, the method further includes: Based on each matching degree, a second variance, a third matching degree, and a fourth matching degree are obtained; the second variance is the variance of each matching degree; the third matching degree is the maximum value among all matching degrees; and the fourth matching degree is the minimum value among all matching degrees. Based on the second variance, the third matching degree, and the fourth matching degree, the correction coefficient is obtained; The sleep risk value is corrected based on the correction factor.

[0014] Secondly, this application proposes a technical solution for a neonatal sleep apnea early warning device, which includes: The reading module is used to acquire initial audio data and monitor audio data of the newborn's sleep; the initial audio data is audio data acquired when the newborn is in a stable and normal breathing state. The processing module is used to acquire a dominant frequency signal based on the initial audio data; the dominant frequency signal is the signal generated by the newborn's breathing frequency during sleep. Furthermore, based on the monitored audio data, multiple time-series data segments are acquired; Furthermore, based on the main frequency signal and each time-series data segment, a matching degree corresponding to each time-series data segment is obtained; the matching degree is at least used to characterize the degree of similarity between the main frequency signal and the corresponding time-series data segment. Furthermore, based on each matching degree, a sleep risk value is obtained; the sleep risk value is used at least to characterize the probability of the newborn experiencing apnea. The early warning module is used to issue an early warning if the sleep risk value is greater than or equal to a first preset value.

[0015] As a specific solution in the technical solution of this application, the processing module is further configured to acquire first audio data; the first audio data is audio data that meets first preset conditions; the first preset conditions include: the first audio data is audio data collected by a microphone from a preset area; the preset area includes an area 10 cm to 30 cm in front of or to the side of the newborn's mouth and nose; And, based on the first audio data, second audio data is obtained; the second audio data is the audio data in the first audio data that meets the second preset condition; And, based on the second audio data, the initial audio data is obtained.

[0016] As a specific solution in this application, the processing module is further configured to acquire multiple first component signals based on the initial audio data; the first component signals are signals obtained from the initial audio data based on empirical mode decomposition. Furthermore, the target signal is obtained by traversing each first component signal; the target signal is any signal in each first component signal for which no corresponding reference degree has been obtained; the reference degree is used at least to characterize the probability that the target signal is the main frequency signal. And, based on the target signal, obtain the reference degree corresponding to the target signal; And, after each first component signal has acquired its corresponding reference degree, the main frequency signal is acquired based on each first component signal; the main frequency signal is the signal with the largest corresponding reference degree among each first component signal.

[0017] As a specific solution in the technical solution of this application, the processing module is further configured to fit the target signal and obtain a fitting curve; Furthermore, based on the fitted curve, a first period and a first variance are obtained; the first period is the minimum positive period of the fitted curve; the first variance is the sum or average of the variances of each maximum value and the variances of each minimum value in the fitted curve. Furthermore, based on the first period, a first difference is obtained; the first difference is equal to the absolute value of the difference between the first period and the second period; the second period is equal to the average value of the minimum positive periods of the fitted curves corresponding to each first component signal. Furthermore, based on the first difference and the first variance, the reference degree corresponding to the target signal is obtained; the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance.

[0018] As a specific solution in the technical solution of this application, the processing module is further configured to traverse each time series data segment and obtain the current data segment; the current data segment is any data segment in each time series data segment that has not obtained the corresponding matching degree; Furthermore, based on the current data segment and the main frequency signal, a first similarity is obtained; the first similarity is the similarity between the current data segment and the main frequency signal. And, based on the first similarity, obtain the matching degree corresponding to the current data segment.

[0019] As a specific solution in this application, the processing module is further configured to obtain a first average value, a second average value, and a second similarity based on the current data segment and the main frequency signal; the first average value is the average value of each extreme point after normalization of the current data segment; the second average value is the average value of each extreme point after normalization of the main frequency signal; the second similarity is negatively correlated with the dynamic time warping distance between the current data segment and the main frequency signal. Furthermore, based on the first average value and the second average value, a second difference value is obtained; the second difference value is equal to the absolute value of the difference between the first average value and the second average value. Furthermore, the first similarity is obtained based on the second difference and the second similarity; the first similarity is negatively correlated with the second difference, and the first similarity is positively correlated with the second similarity.

[0020] As a specific solution in this application, the processing module is further configured to acquire multiple second component signals based on the current data segment; the second component signals are signals obtained from the current data segment based on empirical mode decomposition. Furthermore, multiple signal pairs are obtained based on multiple first component signals and multiple second component signals; each first component signal is a signal obtained from the initial audio data based on empirical mode decomposition; the absolute value of the Pearson correlation coefficient of the first component signal and the second component signal in the signal pair is greater than or equal to a third preset value; Furthermore, based on each signal pair, a first quantity and a second quantity are obtained; the first quantity is the total quantity of each signal pair; the second quantity is the total quantity of each first component signal. Furthermore, based on the first quantity and the second quantity, a quantity ratio is obtained; the quantity ratio is the ratio of the first quantity to the second quantity. Furthermore, based on the first similarity and the quantity ratio, the matching degree corresponding to the current data segment is obtained.

[0021] As a specific solution in the technical solution of this application, the processing module is further configured to obtain a first matching degree and a preset number of second matching degrees based on each matching degree; the timing of each second matching degree and the first matching degree is sequentially adjacent, and the first matching degree is the matching degree corresponding to the last data segment in each timing data segment; Furthermore, each second matching degree is fitted to the first matching degree to obtain a fitted straight line; And, based on the fitted line, the slope of the line is obtained; the slope of the line is the slope of the fitted line; Furthermore, the sleep risk value is obtained based on the slope of the straight line and the first matching degree; the first matching degree is the matching degree corresponding to the last data segment in each time series data segment.

[0022] As a specific solution in the technical solution of this application, the processing module is further configured to obtain a second variance, a third matching degree, and a fourth matching degree based on each matching degree; the second variance is the variance of each matching degree; the third matching degree is the maximum value among all matching degrees; and the fourth matching degree is the minimum value among all matching degrees. Furthermore, a correction coefficient is obtained based on the second variance, the third matching degree, and the fourth matching degree; Furthermore, the sleep risk value is corrected based on the correction coefficient.

[0023] Compared with the prior art, the beneficial effects of this application are: The embodiment of the neonatal sleep apnea early warning method proposed in this application first acquires initial and monitored acoustic data of the newborn's sleep. The initial acoustic data is used to extract the dominant frequency signal representing the individual's respiratory frequency as a personalized benchmark. The monitored acoustic data is then divided into multiple time-series data segments. A matching metric is used to quantify the similarity between the data from each time period and the dominant frequency signal. Finally, a sleep hazard value is calculated based on the matching degree to achieve an early warning. This solution specifically addresses the pain points of existing acoustic monitoring, such as weak respiratory signals, large individual differences, and environmental noise interference. By using a personalized dominant frequency signal as a reference, it avoids the problem of insufficient adaptability of general standards. The combination of time-series segmentation and matching degree analysis can capture the dynamic changes in respiratory rhythm in real time, providing more timely warnings compared to traditional post-event feedback monitoring methods. By quantifying indicators to derive sleep hazard values ​​layer by layer, it effectively distinguishes between apnea and non-respiratory acoustic signals, sleep stage fluctuations, and other interfering factors, significantly reducing false alarms and missed alarms. Furthermore, it requires no invasive operation, making it suitable for long-term continuous home monitoring scenarios, providing reliable protection for the sleep breathing safety of newborns, especially premature infants and low birth weight infants. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a neonatal sleep apnea early warning method proposed in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a neonatal sleep apnea early warning device proposed in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] The terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. For example, the first audio data and the second audio data mentioned below belong to different audio data. It should be understood that such names can be used interchangeably where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules appearing in the embodiments of this application is merely a logical division. In actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms. None of these are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0027] To address the technical problems mentioned in the background section regarding the tendency for false alarms or missed alarms in existing methods for monitoring neonatal sleep apnea based on neonatal audio data, this application proposes an embodiment of a neonatal sleep apnea early warning method. For example... Figure 1 As shown, the neonatal sleep apnea early warning method includes steps 100 to 600.

[0028] Step 100: Obtain initial audio data and monitor audio data of the newborn's sleep.

[0029] In this embodiment, the initial audio data refers to the baseline audio data collected by the microphone that accurately reflects the respiratory rhythm of the newborn during normal sleep. The core purpose of the initial audio data is to extract the dominant frequency signal characterizing the newborn's sleep respiratory frequency, providing a standard reference for comparing subsequent monitoring data. That is, the initial audio data is the audio data acquired when the newborn is in a stable and normal breathing state.

[0030] In this embodiment, the monitored audio data refers to the audio data stream continuously and in real time collected by the same acquisition device (i.e., microphone) during the newborn's sleep. The audio data stream includes breathing-related sound signals, environmental noise and other non-respiratory acoustic signals during the newborn's sleep, which are used to perform similarity matching with the main frequency signal corresponding to the initial audio data, thereby determining whether the breathing state is abnormal.

[0031] In this embodiment, any reasonable method can be used to obtain the initial audio data of the newborn's sleep, as long as the obtained initial audio data of the newborn's sleep can reflect the breathing rhythm of the newborn in a normal sleep state. For example, in this embodiment, obtaining the initial audio data of the newborn's sleep may include steps 110 to 130.

[0032] Step 110: Obtain the first audio data.

[0033] In this embodiment, the first audio data is audio data that meets the first preset conditions.

[0034] In this embodiment, the purpose of setting the first preset condition is to limit the acquisition environment of the first audio data, thereby minimizing interference from environmental noise and non-respiratory acoustic signals and improving the signal-to-noise ratio of the neonatal respiratory-related sound signals in the acquired audio data. In other words, the first preset condition is any condition that allows for obtaining clear respiratory-related sound signals from the newborn. For example, the first preset condition may include the first audio data being audio data acquired by a microphone from a preset area; the preset area includes an area 10 to 30 centimeters in front of or to the side of the newborn's mouth and nose; or, the first preset condition may also include the microphone being kept away from noise sources such as air conditioner vents, humidifiers, and speakers, and maintaining a reasonable distance from non-respiratory sound sources such as the newborn's crying and limb rubbing, to reduce environmental interference; or, the first preset condition may also include the use of a high-sensitivity miniature electret microphone, etc.

[0035] Step 120: Based on the first audio data, obtain the second audio data.

[0036] In this embodiment, the second audio data is the audio data in the first audio data that meets the second preset condition.

[0037] In this embodiment, the purpose of setting the second preset condition is to further refine the first audio data after it has been filtered by the first preset condition. This involves removing invalid data segments containing sudden environmental noise (e.g., sudden operation of household appliances and occasional conversations among family members), non-respiratory-related sound signals from newborns (e.g., crying, limb rubbing, and clothing shaking), and periods of unstable breathing (e.g., early sleep onset and light sleep turning). Only stable, continuous audio segments that accurately reflect the normal sleep-breathing rhythm of newborns are retained. This further improves the signal-to-noise ratio and purity of the initial audio data, providing high-quality basic data for subsequent extraction of the dominant frequency signal based on empirical mode decomposition. It avoids interference from invalid data leading to deviations in dominant frequency signal identification, thereby ensuring the accuracy of subsequent matching degree calculations and sleep hazard value assessments, and reducing the risk of false alarms or missed alarms during the warning process. In other words, in this embodiment, the second preset condition can be that the amplitude fluctuation range of the second audio data is less than a threshold (amplitude variance less than 0.01 within a 5-second or 10-second sliding window; or, short-term energy fluctuation range less than 30%, zero crossover rate coefficient of variation less than 0.2, etc.).

[0038] In this embodiment, the second audio data can be manually extracted from the first audio data. Alternatively, an automated algorithm can be used to automatically filter the first audio data (i.e., extract the segment with the smallest fluctuation in the first audio data as the second audio data). Automated filtering of audio data using algorithms is a mature technology and will not be elaborated upon here. For example, the automated algorithm could be a variance threshold filtering algorithm based on a sliding window. This algorithm iterates through the first audio data using a fixed-length sliding window (e.g., 5 seconds / window), calculates the variance of the audio signal amplitude within each window, and filters out continuous window segments with variances less than a preset threshold. These segments are then spliced ​​together to serve as the second audio data. Alternatively, an adaptive noise suppression combined with a stationarity detection algorithm can be used. First, high-frequency environmental noise is removed using wavelet threshold denoising. Then, based on the short-time energy and zero-crossing rate characteristics of the signal, stationary data segments with short-time energy fluctuations less than 30% and zero-crossing rate variation coefficients less than 0.2 are identified and retained as the second audio data.

[0039] Step 130: Based on the second audio data, obtain the initial audio data.

[0040] In this embodiment, the second audio data can be directly used as the initial audio data. Alternatively, further noise reduction and signal enhancement processing can be applied to the second audio data to further improve data purity. Specifically, adaptive noise suppression algorithms can be used to remove residual environmental noise (e.g., separating high-frequency noise components through wavelet thresholding), or signal amplitude normalization processing can be used to unify the dynamic range of the data.

[0041] Step 200: Based on the initial audio data, obtain the main frequency signal.

[0042] In this embodiment, the dominant frequency signal is the signal generated by the newborn's breathing frequency during sleep. Specifically, step 200 involves obtaining the dominant frequency signal based on the initial audio data, including steps 210 to 240.

[0043] Step 210: Based on the initial audio data, acquire multiple first component signals.

[0044] In this embodiment, the first component signal is the signal obtained from the initial audio data based on empirical mode decomposition.

[0045] It is important to note that Empirical Mode Decomposition (EMD) is an adaptive signal processing method. Its core principle is to decompose complex, nonlinear, and non-stationary signals (e.g., initial audio data containing mixed components such as breathing sounds and environmental noise) into a series of physically meaningful Intrinsic Mode Functions (IMFs) and a residual component, based on the signal's own time-scale characteristics. This decomposition process does not require pre-defined basis functions; it is achieved through iterative selection, and is a mature technology, which will not be elaborated upon here. In other words, in this embodiment, performing EMD on the initial audio data to obtain multiple first-component signals is also a mature technology, and will not be elaborated upon here.

[0046] Step 220: Traverse each first component signal to obtain the target signal.

[0047] In this embodiment, the target signal is any signal among the various first component signals that has not acquired a corresponding reference degree. The reference degree is used at least to characterize the probability that the target signal is a main frequency signal.

[0048] In this embodiment, the dominant frequency signal refers to the signal extracted from the initial audio data of the newborn during normal sleep and that can stably characterize the core features of the newborn's respiratory rhythm. Specifically, it is the dominant acoustic signal component that is directly related to the newborn's respiratory frequency during sleep.

[0049] In this embodiment, the purpose of traversing each first component signal to obtain the target signal is to assign an independent reference degree calculation process to each unevaluated component signal by analyzing each first component signal obtained from empirical mode decomposition, ensuring that no potential signal that may characterize the respiratory frequency of a newborn is overlooked. Since the initial audio data, after empirical mode decomposition, will yield various types of first component signals, including the respiratory dominant frequency signal, environmental noise components, and non-respiratory acoustic signal components, and the physical meaning of each component signal is unclear, making it impossible to directly determine which signal corresponds to the respiratory dominant frequency, the traversal method is used to select the target signal for which no reference degree has been obtained and calculate its reference degree. This systematically quantifies the probability of each component signal matching the dominant frequency signal, providing a comprehensive and accurate basis for subsequently selecting the dominant frequency signal with the highest reference degree. This avoids deviations in dominant frequency signal extraction caused by direct subjective judgment or the omission of some signals, thereby ensuring the accuracy of subsequent matching degree calculations and sleep apnea warnings.

[0050] In this embodiment, steps 230 to 240 are performed for each target signal until each target signal obtains a corresponding reference degree.

[0051] Step 230: Based on the target signal, obtain the reference degree corresponding to the target signal.

[0052] In this embodiment, any reasonable method can be used to obtain the reference degree of the target signal, as long as the obtained reference degree is close to the probability that the target signal is the main frequency signal.

[0053] It should be noted that, compared to the signals generated during neonatal sleep breathing, high-frequency noise produces signals with shorter periods, while low-frequency noise produces signals with longer periods, and the extreme values ​​of both high-frequency and low-frequency noise fluctuate significantly. Therefore, in a specific embodiment of this application, obtaining the reference level corresponding to the target signal may include steps 231 to 234.

[0054] Step 231: Fit the target signal to obtain a fitting curve.

[0055] It is important to understand that multinomial fitting of signals to obtain corresponding fitting curves is a mature technique, which will not be elaborated upon here.

[0056] Step 232: Based on the fitted curve, obtain the first period and the first variance.

[0057] In this embodiment, the first period is the smallest positive period of the fitted curve. The first variance is the sum or average of the variances of the maxima and minima of the fitted curve.

[0058] Step 233: Based on the first period, obtain the first difference.

[0059] In this embodiment, the first difference is equal to the absolute value of the difference between the first period and the second period; the second period is equal to the average value of the minimum positive period of the fitting curves corresponding to each first component signal. Of course, the second period can also be a preset empirical value of the normal breathing cycle of a newborn (such as 1.5 seconds).

[0060] In this embodiment, if the target signal is a main frequency signal, then both the first difference and the first variance are relatively small. In other words, the smaller the first difference and the first variance corresponding to a target signal, the more likely that signal is to be a main frequency signal.

[0061] Step 234: Based on the first difference and the first variance, obtain the reference degree corresponding to the target signal.

[0062] In this embodiment, the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance. That is, in this embodiment, any reasonable method can be used to obtain the reference degree corresponding to the target signal based on the first difference and the first variance, as long as the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance. For example, in one embodiment of this application, step 234, the formula for calculating the reference degree corresponding to the target signal based on the first difference and the first variance, can be as follows: in, Indicates the reference degree corresponding to the target signal; Indicates the first difference; Indicates the first variance; This represents the zero-prevention coefficient, used to avoid a denominator of 0. The zero-prevention coefficient can be any positive integer, such as 0.1 or 1. In this embodiment, the higher the reference level, the greater the probability that the target signal is the main frequency signal; conversely, the lower the reference level, the lower the probability that the target signal is the main frequency signal.

[0063] It should be clear that, in the embodiments of this application, in order to avoid the denominator being zero, when calculating various ratios (e.g., quantity ratios below), a zero-prevention coefficient similar to that mentioned above can be added to the denominator, which will not be listed and elaborated on in detail below.

[0064] In another embodiment of this application, step 234, based on the first difference and the first variance, the formula for calculating the reference degree corresponding to the target signal can be as follows: in, Indicates the reference degree corresponding to the target signal; Indicates the first difference; Indicates the first variance; This represents an exponential function with the natural constant e as its base. In this embodiment, the greater the reference degree, the greater the probability that the target signal is the main frequency signal; conversely, the smaller the reference degree, the smaller the probability that the target signal is the main frequency signal.

[0065] Step 240: After each first component signal has acquired its corresponding reference degree, the main frequency signal is acquired based on each first component signal.

[0066] In this embodiment, the main frequency signal is the signal with the highest reference degree among all the first component signals.

[0067] Step 300: Based on the monitored audio data, acquire multiple time-series data segments.

[0068] In this embodiment, multiple time-series data segments can be obtained based on the monitored audio data using any reasonable method. For example, the monitored audio data can be evenly divided into multiple time-series data segments, and the time length of each time-series data segment can be equal to a second preset value.

[0069] In this embodiment, the second preset value can be set according to requirements, for example, the second preset value can be 5 minutes or 10 minutes, etc.

[0070] In order to ensure that each time-series data segment contains multiple complete respiratory cycle data of the newborn, so as to facilitate the subsequent comparison of the similarity between the time-series data segment and the main frequency signal, in one embodiment of this application, step 300, based on the monitored audio data, to obtain multiple time-series data segments may include steps 310 and 320.

[0071] Step 310: Obtain the data segment length.

[0072] In this embodiment, the data segment length is equal to a positive integer multiple of the neonatal respiratory cycle. For example, if the neonatal respiratory cycle length is 3 seconds, the data segment length can be equal to 3 seconds × 3 = 9 seconds, 3 seconds × 5 = 15 seconds, or 3 seconds × 10 = 30 seconds, etc. Selecting positive integer multiples such as 3, 5, and 10 is to ensure that each time-series data segment contains a sufficient number of complete respiratory cycles. This avoids insufficient respiratory feature extraction due to excessively short data segments, and also prevents excessive environmental noise or respiratory state changes from interfering with excessively long data segments, thereby improving the accuracy of subsequent matching with the main frequency signal. The specific multiple can be adjusted according to actual monitoring accuracy requirements; no further restrictions are imposed here.

[0073] Step 320: Divide the monitored audio data into multiple time-series data segments according to the data segment length.

[0074] Dividing audio data (i.e., monitoring audio data) into multiple data segments (i.e., various time-series data segments) according to time length is a mature technology, which will not be elaborated here.

[0075] Step 400: Based on the main frequency signal and each time data segment, obtain the matching degree corresponding to each time data segment.

[0076] In this embodiment, the matching degree is used at least to characterize the degree of similarity between the main frequency signal and the corresponding time-series data segment.

[0077] As mentioned above, the dominant frequency signal refers to the signal extracted from the initial audio data of a newborn during normal sleep, which can stably characterize the core features of the newborn's respiratory rhythm. The greater the similarity between a time-series data segment and the dominant frequency signal, the more normal the newborn's sleep breathing is within the corresponding time period; conversely, the smaller the similarity, the more abnormal the newborn's sleep breathing is within the corresponding time period. In other words, in this embodiment, the similarity between the dominant frequency signal and each time-series data segment (e.g., the dynamic time warping distance mentioned below) can be directly used as the matching degree for a one-to-one correspondence with each time-series data segment.

[0078] It is important to note that apnea monitoring requires real-time performance. Apnea itself is characterized by a short-term cessation of breathing, and the monitoring scenario is susceptible to interference from various factors. On the one hand, newborns' movements such as turning over or turning their heads during sleep may cause them to move away from the monitoring device (e.g., the microphone mentioned above), weakening the collected exhalation sounds, which can easily be misjudged as apnea. On the other hand, human sleep has different stages, and the respiratory frequency and rhythm of each stage have natural differences. Furthermore, the intervention of environmental noise can cause interference and aliasing of the collected monitoring audio data, making it difficult to accurately capture the true breathing pattern. Therefore, if the similarity between the main frequency signal and each time-series data segment is used as the matching degree for judgment, it is difficult to accurately distinguish between the above-mentioned interference factors and the true apnea situation, which can easily lead to false alarms. Based on this, in one embodiment of this application, step 400, based on the main frequency signal and each time-series data segment, obtains the matching degree corresponding to each time-series data segment, including steps 410 to 430.

[0079] Step 410: Traverse each time series data segment and obtain the current data segment.

[0080] In this embodiment, the current data segment is any data segment in each time-series data segment that has not obtained the corresponding matching degree.

[0081] In this embodiment, the purpose of traversing each time-series data segment to obtain the current data segment is to: ensure that each continuous monitoring period covered by the monitored audio data is analyzed one by one without omission, avoiding the loss of respiratory status information during key monitoring periods due to batch processing or random selection of data segments; for each current data segment for which the matching degree has not been calculated, the similarity comparison and matching degree calculation with the main frequency signal are carried out independently, adapting to the possible dynamic changes in the respiratory state during the newborn's sleep, and realizing the accurate capture of respiratory characteristics at different time periods; at the same time, by processing them one by one, combined with the subsequent multi-level matching degree calculation logic based on extreme value features and component signal correlation, the influence of accidental interference factors such as environmental noise and limb micro-movements in a single data segment is effectively filtered out, ensuring that the matching degree corresponding to each time-series data segment can truly reflect the degree of fit between the respiratory signal and the normal respiratory main frequency signal during that period, providing comprehensive and reliable data support for subsequent calculation of sleep risk values ​​based on continuous matching degree sequences and realizing accurate early warning.

[0082] In this embodiment, steps 420 to 430 are executed for each current data segment until each time-series data segment obtains a corresponding matching degree.

[0083] Step 420: Based on the current data segment and the main frequency signal, obtain the first similarity.

[0084] In this embodiment, the first similarity is the similarity between the current data segment and the main frequency signal.

[0085] It is important to understand that obtaining the similarity between two data segments (i.e., the current data segment and the main frequency signal) is a mature technology. For example, the dynamic time warping distance between the current data segment and the main frequency signal can be calculated as the first similarity; or, the Pearson correlation coefficient between the current data segment and the main frequency signal can be calculated and normalized, and the normalized correlation coefficient can be used as the first similarity.

[0086] As mentioned above, newborns may experience natural fluctuations in their respiratory rhythm due to sleep stage transitions during sleep. Furthermore, environmental noise and minor limb movements can easily cause local distortions in signal amplitude or temporal pattern. Therefore, relying solely on a single-dimensional similarity index (e.g., dynamic time regularization distance or Pearson correlation coefficient) is insufficient to accurately distinguish between "true respiratory abnormalities" and "interference / normal fluctuations," potentially leading to bias in the first similarity assessment. Based on this, in a specific embodiment of this application, step 420, based on the current data segment and the dominant frequency signal, obtains the first similarity, including steps 421 to 423.

[0087] Step 421: Based on the current data segment and the main frequency signal, obtain the first average value, the second average value, and the second similarity.

[0088] In this embodiment, the first average value is the average value of all extreme points after normalization of the current data segment. The second average value is the average value of all extreme points after normalization of the main frequency signal. The second similarity is negatively correlated with the dynamic time warping distance between the current data segment and the main frequency signal.

[0089] Step 422: Obtain the second difference based on the first average value and the second average value.

[0090] In this embodiment, the second difference is equal to the absolute value of the difference between the first average value and the second average value. In this embodiment, the more similar the current data segment and the main frequency signal are, the smaller the second difference is; the less similar the current data segment and the main frequency signal are, the larger the second difference is.

[0091] Step 423: Obtain the first similarity based on the second difference and the second similarity.

[0092] In this embodiment, the first similarity is negatively correlated with the second difference, and the first similarity is positively correlated with the second similarity. That is, in this embodiment, any reasonable method can be used to obtain the first similarity based on the second difference and the second similarity, as long as the first similarity is negatively correlated with the second difference and positively correlated with the second similarity. For example, in one embodiment of this application, step 423, the formula for calculating the first similarity based on the second difference and the second similarity, can be as follows: in, Indicates the first similarity; Indicates the second difference; Indicates the second similarity; This represents the zero-prevention coefficient, used to avoid a denominator of 0. The zero-prevention coefficient can be any positive integer, such as 0.1 or 1. In this embodiment, the smaller the second difference, the greater the first similarity; the larger the second difference, the smaller the first similarity; the greater the second similarity, the greater the first similarity; and the smaller the second similarity, the smaller the first similarity.

[0093] In another embodiment of this application, step 423, based on the second difference and the second similarity, the formula for calculating the first similarity can be as follows: in, Indicates the first similarity; Indicates the second difference; Indicates the second similarity; This represents an exponential function with the natural constant e as its base. This represents the zero-prevention coefficient, used to avoid a denominator of 0. The zero-prevention coefficient can be any positive integer, such as 0.1 or 1. In this embodiment, the smaller the second difference, the greater the first similarity; the larger the second difference, the smaller the first similarity; the greater the second similarity, the greater the first similarity; and the smaller the second similarity, the smaller the first similarity.

[0094] Step 430: Based on the first similarity, obtain the matching degree corresponding to the current data segment.

[0095] In this embodiment, the first similarity can be directly used as the matching degree.

[0096] It should be noted that if a newborn experiences apnea, the corresponding current data segment will definitely lack a large number of signals related to the newborn's breathing. In order to obtain a matching degree that can more accurately characterize whether a newborn has apnea, in one embodiment of this application, step 430, based on the first similarity, obtains the matching degree corresponding to the current data segment, which may include steps 431 to 435.

[0097] Step 431: Based on the current data segment, acquire multiple second component signals.

[0098] In this embodiment, the second component signal is the signal obtained by the current data segment based on empirical mode decomposition.

[0099] Step 432: Based on multiple first component signals and multiple second component signals, obtain multiple signal pairs.

[0100] In this embodiment, each first component signal is a signal obtained from the initial audio data based on empirical mode decomposition. The absolute value of the Pearson correlation coefficient between the first component signal and the second component signal in the signal pair is greater than or equal to a third preset value. That is, in this embodiment, if the absolute value of the Pearson correlation coefficient between a certain second component signal and any first component signal that does not form a signal pair is greater than or equal to the third preset value, then the second component signal and the first component signal can form a signal pair.

[0101] In this embodiment, a third preset value can be set according to needs or experience. For example, the third preset value can be 0.7 or 0.8, etc.

[0102] As mentioned above, since the initial audio data was collected in a quiet environment within a preset area 10 to 30 centimeters directly in front of the newborn's mouth and nose, and has undergone environmental screening, invalid data removal, and noise reduction processing, its purity and signal-to-noise ratio are extremely high. Therefore, most of the signals in each first component signal (obtained from the initial audio data through empirical mode decomposition) are effective signals directly related to the newborn's breathing, and can accurately reflect the core characteristics of normal breathing such as rhythm and amplitude. In other words, the fewer signal pairs formed by each first component signal and each second component signal (obtained from the current data segment through empirical mode decomposition), the fewer signal components related to normal neonatal breathing are contained in the current data segment. It is more affected by non-respiratory interference signals such as environmental noise, limb friction, and clothing shaking. Alternatively, the neonatal respiratory signal itself may be weakened, disordered, or even missing within the time period corresponding to the current data segment. This means that the current data segment is less consistent with the normal respiratory frequency signal, and the matching degree calculated later will be smaller. This indicates that the probability of the neonatal respiratory rhythm abnormality (e.g., apnea or weakened breathing) during this period is higher.

[0103] Step 433: Based on each signal pair, obtain the first quantity and the second quantity.

[0104] In this embodiment, the first quantity is the total number of each signal pair. The second quantity is the total number of each first component signal.

[0105] Step 434: Obtain the quantity ratio based on the first quantity and the second quantity.

[0106] In this embodiment, the quantity ratio is the ratio of the first quantity to the second quantity.

[0107] Step 435: Based on the first similarity and the quantity ratio, obtain the matching degree corresponding to the current data segment.

[0108] In this embodiment, the larger the ratio of the first similarity to the number, the higher the degree of fit between the current data segment and the initial audio data of the newborn in normal sleep, that is, the higher the consistency between the newborn's breathing rhythm and the normal sleep breathing rhythm, the lower the probability of weakened, disordered, or apnea, and the greater the corresponding matching degree. Conversely, the smaller the ratio of the first similarity to the number, the lower the degree of fit between the current data segment and the initial audio data of the newborn in normal sleep, that is, the lower the consistency between the newborn's breathing rhythm and the normal sleep breathing rhythm, the higher the probability of weakened, disordered, or apnea, and the smaller the corresponding matching degree. In other words, in this embodiment, the matching degree corresponding to the current data segment only needs to be positively correlated with the first similarity and the number ratio. For example, in one embodiment of this application, step 435, the formula for calculating the matching degree corresponding to the current data segment based on the first similarity and the number ratio, can be as follows: in, Indicates the degree of matching; Indicates the first similarity; Indicates the ratio of quantities; This represents an exponential function with the natural constant e as its base. This represents the zero-prevention coefficient, used to avoid the denominator being 0. The zero-prevention coefficient can be any positive integer, such as 0.1 or 1.

[0109] In another embodiment of this application, step 435, the formula for calculating the matching degree corresponding to the current data segment based on the first similarity and the quantity ratio, can be as follows: in, Indicates the degree of matching; Indicates the first similarity; It represents the ratio of quantities.

[0110] Step 500: Obtain sleep risk values ​​based on each matching degree.

[0111] In this embodiment, the sleep hazard value is used at least to characterize the probability of the newborn experiencing apnea.

[0112] As mentioned above, a higher matching degree indicates a greater consistency between the newborn's respiratory rhythm and the normal sleep respiratory rhythm, and a lower probability of the newborn experiencing weakened, disordered, or apnea. Therefore, in this embodiment, any reasonable method can be used to obtain the sleep risk value based on each matching degree. For example, the sleep risk value can be the average of several matching degrees (e.g., 10 or 20) closest to the current time; or, step 500, obtaining the sleep risk value based on each matching degree, includes steps 510 to 540.

[0113] Step 510: Based on each matching degree, obtain the first matching degree and a preset number (such as 10 or 20) of second matching degrees.

[0114] In this embodiment, the timing of each second matching degree and the first matching degree is sequentially adjacent, and the first matching degree is the matching degree corresponding to the last data segment in each timing data segment.

[0115] Step 520: Fit each second matching degree to the first matching degree to obtain a fitted straight line.

[0116] It is important to understand that fitting multiple values ​​(i.e., each second matching degree and the first matching degree) to obtain a fitted line is a mature technique. For example, it can be achieved through the least squares method or gradient descent method, which will not be elaborated on here.

[0117] Step 530: Based on the fitted line, obtain the slope of the line.

[0118] In this embodiment, the slope of the straight line is the slope of the fitted straight line.

[0119] In this embodiment, since the matching degree directly characterizes the degree of fit between the respiratory signal and the normal respiratory frequency signal during the monitoring period, and the matching degree is positively correlated with the normality of the newborn's respiratory rhythm, if the newborn shows signs of apnea or has already experienced apnea, a large number of respiratory-related feature signals will be missing from the monitored audio data, which will lead to a continuous and sharp downward trend in the matching degree of each time series data segment. This is reflected in the fitted line as a negative slope, that is, a small slope. Alternatively, if the newborn's breathing is intermittent, this will be reflected in the fitted line as a sudden positive or negative slope. Conversely, if the newborn's respiratory state is stable, the matching degree will remain at a high level and fluctuate gently, that is, the slope of the fitted line will be a positive value approaching 0, indicating that the respiratory rhythm fits the normal state stably.

[0120] Step 540: Obtain the sleep risk value based on the slope of the straight line and the first matching degree.

[0121] In this embodiment, the first matching degree is the matching degree corresponding to the last data segment in each time series data segment.

[0122] In this embodiment, the sleep risk value can be obtained based on the slope of the straight line and the first matching degree in any reasonable manner. For example, in one embodiment of this application, step 540, the calculation formula for obtaining the sleep risk value based on the slope of the straight line and the first matching degree, can be as follows: in, Indicates the risk value of sleep; Indicates the slope of a straight line; This indicates the absolute value; Indicates the first degree of match; This represents the zero-prevention coefficient, used to avoid a denominator of 0. The zero-prevention coefficient can be any positive integer, such as 0.1 or 1. In this embodiment, the smaller the slope of the straight line and the first matching degree, the greater the sleep risk value; conversely, the greater the slope of the straight line and the first matching degree, the smaller the sleep risk value.

[0123] In another embodiment of this application, step 540, based on the slope of the straight line and the first matching degree, the calculation formula for obtaining the sleep risk value can be as follows: in, Indicates the risk value of sleep; Indicates the slope of a straight line; Indicates the first degree of match; This represents an exponential function with the natural constant e as its base. In this embodiment, the smaller the slope of the straight line and the first degree of matching, the greater the sleep risk value; conversely, the greater the slope of the straight line and the first degree of matching, the smaller the sleep risk value.

[0124] It is important to note that the sleep risk value calculated solely based on the slope of the straight line and the first matching degree does not fully consider the overall fluctuation characteristics and extreme values ​​of the matching degree sequence. On the one hand, occasional disturbances such as brief limb movements or sudden environmental noises may occur during a newborn's sleep, leading to abnormally high or low extreme values ​​in individual matching degrees. Directly incorporating these into the calculation can easily result in misjudgment of the sleep risk value. On the other hand, the variance of the matching degree reflects the stability of the respiratory state. The larger the variance, the more drastic the fluctuation in the degree of fit between the respiratory signal and the normal dominant frequency signal, which may be a precursor to respiratory rhythm disorder. This crucial information cannot be captured by relying solely on the slope and the latest matching degree. Based on this, in one embodiment of this application, after obtaining the sleep risk value based on the slope of the straight line and the first matching degree in step 530, the method further includes steps 540 to 560.

[0125] Step 540: Based on each matching degree, obtain the second variance, the third matching degree, and the fourth matching degree.

[0126] In this embodiment, the second variance is the variance of each matching degree. The third matching degree is the maximum value among all matching degrees. The fourth matching degree is the minimum value among all matching degrees.

[0127] Step 550: Obtain the correction coefficient based on the second variance, the third matching degree, and the fourth matching degree.

[0128] Specifically, the formula for calculating the correction factor is as follows: in, Indicates the correction factor; These are the preset weighting coefficients for the second variance; Indicates the second variance; for The preset weighting coefficients; Indicates the third degree of matching; Indicates the fourth degree of matching; This indicates that the absolute value is being calculated.

[0129] It should be noted that the weighting in the quantification of the correction coefficient for sleep risk values ​​is based on an in-depth analysis of the correlation between respiratory signal fluctuation characteristics and clinical risk. Second variance Reflecting the overall dispersion of the matching sequence, it can continuously capture changes in the stability of respiratory rhythm, and has higher sensitivity and early warning value for progressive and persistent respiratory disturbances; range It mainly reflects the degree of a single extreme deviation and is easily affected by instantaneous interference. Although it has a certain indicative role in sudden anomalies, its stability is relatively weak.

[0130] To balance the contributions of both in risk assessment, this embodiment, after validation with multiple sets of clinical data and parameter optimization, sets the preset weighting coefficient for the second variance to 0.6 and the preset weighting coefficient for the range to 0.4. This ratio aims to prioritize the identification of systematic fluctuations in respiratory rhythm while also taking into account occasional large deviations, thereby enhancing the robustness and clinical interpretability of sleep risk values ​​in real-world disturbance environments.

[0131] Step 560: Correct the sleep risk value based on the correction coefficient.

[0132] Specifically, the revised formula for calculating the sleep risk value can be as follows: in, This indicates the revised sleep risk value; This indicates the sleep risk value mentioned above (i.e., the sleep risk value before the correction). This represents the correction factor.

[0133] Step 600: If the sleep risk value is greater than or equal to the first preset value, an early warning is issued.

[0134] In this embodiment, the first preset value can be set according to requirements, for example, the first preset value can be 5 or 10, etc.

[0135] In this embodiment, any reasonable method can be used for early warning, such as sound warning (e.g., buzzer or mobile phone), light warning (e.g., flashing light), or vibration warning (e.g., mobile phone).

[0136] The embodiment of the neonatal sleep apnea early warning method proposed in this application first acquires initial and monitored acoustic data of the newborn's sleep. The initial acoustic data is used to extract the dominant frequency signal representing the individual's respiratory frequency as a personalized benchmark. The monitored acoustic data is then divided into multiple time-series data segments. A matching metric is used to quantify the similarity between the data from each time period and the dominant frequency signal. Finally, a sleep hazard value is calculated based on the matching degree to achieve an early warning. This solution specifically addresses the pain points of existing acoustic monitoring, such as weak respiratory signals, large individual differences, and environmental noise interference. By using a personalized dominant frequency signal as a reference, it avoids the problem of insufficient adaptability of general standards. The combination of time-series segmentation and matching degree analysis can capture the dynamic changes in respiratory rhythm in real time, providing more timely warnings compared to traditional post-event feedback monitoring methods. By quantifying indicators to derive sleep hazard values ​​layer by layer, it effectively distinguishes between apnea and non-respiratory acoustic signals, sleep stage fluctuations, and other interfering factors, significantly reducing false alarms and missed alarms. Furthermore, it requires no invasive operation, making it suitable for long-term continuous home monitoring scenarios, providing reliable protection for the sleep breathing safety of newborns, especially premature infants and low birth weight infants.

[0137] Having described the neonatal sleep apnea early warning method proposed in the embodiments of this application, the following describes an embodiment of a neonatal sleep apnea early warning device proposed in this application. Figure 2 As shown, the neonatal sleep apnea warning device 10 includes: The reading module 11 is used to acquire initial audio data and monitoring audio data of the newborn's sleep; the initial audio data is audio data acquired when the newborn is in a stable and normal breathing state. Processing module 12 is used to acquire a dominant frequency signal based on the initial audio data; the dominant frequency signal is the signal generated by the breathing frequency of the newborn during sleep. Furthermore, based on the monitored audio data, multiple time-series data segments are acquired; Furthermore, based on the main frequency signal and each time-series data segment, a matching degree corresponding to each time-series data segment is obtained; the matching degree is at least used to characterize the degree of similarity between the main frequency signal and the corresponding time-series data segment. Furthermore, based on each matching degree, a sleep risk value is obtained; the sleep risk value is used at least to characterize the probability of the newborn experiencing apnea. The early warning module 13 is used to issue an early warning if the sleep risk value is greater than or equal to a first preset value.

[0138] As a specific embodiment of this application, the processing module 12 is further configured to acquire first audio data; the first audio data is audio data that meets first preset conditions; the first preset conditions include: the first audio data is audio data collected by a microphone from a preset area; the preset area includes an area 10 cm to 30 cm in front of or to the side of the newborn's mouth and nose; And, based on the first audio data, second audio data is obtained; the second audio data is the audio data in the first audio data that meets the second preset condition; And, based on the second audio data, the initial audio data is obtained.

[0139] As a specific embodiment of this application, the processing module 12 is further configured to acquire a plurality of first component signals based on the initial audio data; the first component signals are signals obtained from the initial audio data based on empirical mode decomposition. Furthermore, the target signal is obtained by traversing each first component signal; the target signal is any signal in each first component signal for which no corresponding reference degree has been obtained; the reference degree is used at least to characterize the probability that the target signal is the main frequency signal. And, based on the target signal, obtain the reference degree corresponding to the target signal; And, after each first component signal has acquired its corresponding reference degree, the main frequency signal is acquired based on each first component signal; the main frequency signal is the signal with the largest corresponding reference degree among each first component signal.

[0140] As a specific embodiment of this application, the processing module 12 is further configured to fit the target signal and obtain a fitting curve; Furthermore, based on the fitted curve, a first period and a first variance are obtained; the first period is the minimum positive period of the fitted curve; the first variance is the sum or average of the variances of each maximum value and the variances of each minimum value in the fitted curve. Furthermore, based on the first period, a first difference is obtained; the first difference is equal to the absolute value of the difference between the first period and the second period; the second period is equal to the average value of the minimum positive periods of the fitted curves corresponding to each first component signal. Furthermore, based on the first difference and the first variance, the reference degree corresponding to the target signal is obtained; the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance.

[0141] As a specific embodiment of this application, the processing module 12 is further configured to traverse each time series data segment and obtain the current data segment; the current data segment is any data segment in each time series data segment that has not obtained the corresponding matching degree; Furthermore, based on the current data segment and the main frequency signal, a first similarity is obtained; the first similarity is the similarity between the current data segment and the main frequency signal. And, based on the first similarity, obtain the matching degree corresponding to the current data segment.

[0142] As a specific embodiment of this application, the processing module 12 is further configured to obtain a first average value, a second average value, and a second similarity based on the current data segment and the main frequency signal; the first average value is the average value of each extreme point after normalization of the current data segment; the second average value is the average value of each extreme point after normalization of the main frequency signal; the second similarity is negatively correlated with the dynamic time warping distance between the current data segment and the main frequency signal; Furthermore, based on the first average value and the second average value, a second difference value is obtained; the second difference value is equal to the absolute value of the difference between the first average value and the second average value. Furthermore, the first similarity is obtained based on the second difference and the second similarity; the first similarity is negatively correlated with the second difference, and the first similarity is positively correlated with the second similarity.

[0143] As a specific embodiment of this application, the processing module 12 is further configured to acquire a plurality of second component signals based on the current data segment; the second component signals are signals obtained from the current data segment based on empirical mode decomposition; Furthermore, multiple signal pairs are obtained based on multiple first component signals and multiple second component signals; each first component signal is a signal obtained from the initial audio data based on empirical mode decomposition; the absolute value of the Pearson correlation coefficient of the first component signal and the second component signal in the signal pair is greater than or equal to a third preset value; Furthermore, based on each signal pair, a first quantity and a second quantity are obtained; the first quantity is the total quantity of each signal pair; the second quantity is the total quantity of each first component signal. Furthermore, based on the first quantity and the second quantity, a quantity ratio is obtained; the quantity ratio is the ratio of the first quantity to the second quantity. Furthermore, based on the first similarity and the quantity ratio, the matching degree corresponding to the current data segment is obtained.

[0144] As a specific embodiment of this application, the processing module 12 is further configured to obtain a first matching degree and a preset number of second matching degrees based on each matching degree; the timing of each second matching degree and the first matching degree is sequentially adjacent, and the first matching degree is the matching degree corresponding to the last data segment in each timing data segment; Furthermore, each second matching degree is fitted to the first matching degree to obtain a fitted straight line; And, based on the fitted line, the slope of the line is obtained; the slope of the line is the slope of the fitted line; Furthermore, the sleep risk value is obtained based on the slope of the straight line and the first matching degree; the first matching degree is the matching degree corresponding to the last data segment in each time series data segment.

[0145] As a specific embodiment of this application, the processing module 12 is further configured to obtain a second variance, a third matching degree, and a fourth matching degree based on each matching degree; the second variance is the variance of each matching degree; the third matching degree is the maximum value among all matching degrees; and the fourth matching degree is the minimum value among all matching degrees. Furthermore, a correction coefficient is obtained based on the second variance, the third matching degree, and the fourth matching degree; Furthermore, the sleep risk value is corrected based on the correction coefficient.

[0146] The embodiment of the neonatal sleep apnea early warning device proposed in this application first acquires initial and monitored acoustic data of the newborn's sleep. The initial acoustic data is used to extract the dominant frequency signal representing the individual's respiratory frequency as a personalized benchmark. The monitored acoustic data is then divided into multiple time-series data segments. A matching metric is used to quantify the similarity between the data from each time period and the dominant frequency signal. Finally, a sleep hazard value is calculated based on the matching degree to achieve an early warning. This solution specifically addresses the pain points of existing acoustic monitoring, such as weak respiratory signals, large individual differences, and environmental noise interference. By using a personalized dominant frequency signal as a reference, it avoids the problem of insufficient adaptability of general standards. The combination of time-series segmentation and matching degree analysis can capture the dynamic changes in respiratory rhythm in real time, providing more timely warnings compared to traditional post-event feedback monitoring methods. By quantifying indicators to derive sleep hazard values ​​layer by layer, it effectively distinguishes between apnea and non-respiratory acoustic signals, sleep stage fluctuations, and other interfering factors, significantly reducing false alarms and missed alarms. Furthermore, it requires no invasive operation, making it suitable for long-term continuous home monitoring scenarios, providing reliable protection for the sleep breathing safety of newborns, especially premature infants and low birth weight infants.

[0147] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0148] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the methods, apparatuses, and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0149] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or modules may be electrical, mechanical, or other forms.

[0150] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0151] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0152] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0153] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video optical disc), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0154] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles of this application.

Claims

1. A method for early warning of sleep apnea in newborns, characterized in that, include: Acquire initial and monitor audio data of newborn sleep; The initial audio data is audio data acquired when the newborn is in a stable and normal breathing state; Based on the initial audio data, a dominant frequency signal is obtained; the dominant frequency signal is the signal generated by the breathing frequency of the newborn during sleep. Based on the monitored audio data, multiple time-series data segments are obtained; Based on the main frequency signal and each time-series data segment, a matching degree corresponding to each time-series data segment is obtained; the matching degree is at least used to characterize the degree of similarity between the main frequency signal and the corresponding time-series data segment. Based on each matching degree, obtain the sleep risk value; The sleep risk value is used to characterize at least the probability of the newborn experiencing apnea; If the sleep risk value is greater than or equal to the first preset value, an early warning will be issued.

2. The neonatal sleep apnea early warning method according to claim 1, characterized in that, The acquisition of initial audio data of newborn sleep includes: Acquire first audio data; the first audio data is audio data that meets first preset conditions; the first preset conditions include: the first audio data is audio data collected by a microphone from a preset area; the preset area includes an area 10 cm to 30 cm in front of or to the side of the newborn's mouth and nose; Based on the first audio data, second audio data is obtained; the second audio data is the audio data in the first audio data that meets the second preset condition; the second preset condition includes: the amplitude fluctuation range of the second audio data is less than the fluctuation threshold; Based on the second audio data, the initial audio data is obtained.

3. The neonatal sleep apnea early warning method according to claim 1, characterized in that, The step of obtaining the main frequency signal based on the initial audio data includes: Based on the initial audio data, multiple first component signals are obtained; the first component signals are signals obtained from the initial audio data based on empirical mode decomposition. The target signal is obtained by traversing each first component signal; the target signal is any signal in each first component signal for which no corresponding reference degree has been obtained; the reference degree is used at least to characterize the probability that the target signal is the main frequency signal. Based on the target signal, obtain the reference degree corresponding to the target signal; After each first component signal has acquired its corresponding reference degree, the main frequency signal is acquired based on each first component signal; the main frequency signal is the signal with the highest corresponding reference degree among each first component signal.

4. The neonatal sleep apnea early warning method according to claim 3, characterized in that, The step of obtaining the reference degree corresponding to the target signal includes: The target signal is fitted to obtain a fitting curve; Based on the fitted curve, a first period and a first variance are obtained; the first period is the minimum positive period of the fitted curve; the first variance is the sum or average of the variances of each maximum value and the variances of each minimum value in the fitted curve. Based on the first period, a first difference is obtained; the first difference is equal to the absolute value of the difference between the first period and the second period; the second period is equal to the average value of the minimum positive period of the fitting curves corresponding to each first component signal. Based on the first difference and the first variance, the reference degree corresponding to the target signal is obtained; the reference degree corresponding to the target signal is negatively correlated with the first difference and the first variance.

5. The neonatal sleep apnea early warning method according to any one of claims 1 to 4, characterized in that, The step of obtaining the matching degree corresponding one-to-one with each time-series data segment based on the main frequency signal and each time-series data segment includes: Traverse each time series data segment to obtain the current data segment; the current data segment is any data segment in each time series data segment for which no corresponding matching degree has been obtained; Based on the current data segment and the main frequency signal, a first similarity is obtained; the first similarity is the similarity between the current data segment and the main frequency signal. Based on the first similarity, the matching degree corresponding to the current data segment is obtained.

6. The neonatal sleep apnea early warning method according to claim 5, characterized in that, The step of obtaining a first similarity based on the current data segment and the main frequency signal includes: Based on the current data segment and the main frequency signal, a first average value, a second average value, and a second similarity are obtained; the first average value is the average value of each extreme point after normalization of the current data segment; the second average value is the average value of each extreme point after normalization of the main frequency signal; the second similarity is negatively correlated with the dynamic time warping distance between the current data segment and the main frequency signal. A second difference is obtained based on the first average and the second average; the second difference is equal to the absolute value of the difference between the first average and the second average. The first similarity is obtained based on the second difference and the second similarity; the first similarity is negatively correlated with the second difference, and the first similarity is positively correlated with the second similarity.

7. The neonatal sleep apnea early warning method according to claim 5, characterized in that, The step of obtaining the matching degree corresponding to the current data segment based on the first similarity includes: Based on the current data segment, multiple second component signals are obtained; the second component signals are signals obtained from the current data segment based on empirical mode decomposition. Multiple signal pairs are obtained based on multiple first component signals and multiple second component signals; each first component signal is a signal obtained from the initial audio data based on empirical mode decomposition; the absolute value of the Pearson correlation coefficient of the first component signal and the second component signal in the signal pair is greater than or equal to a third preset value; Based on each signal pair, a first quantity and a second quantity are obtained; the first quantity is the total quantity of each signal pair; the second quantity is the total quantity of each first component signal. Based on the first quantity and the second quantity, a quantity ratio is obtained; the quantity ratio is the ratio of the first quantity to the second quantity. Based on the first similarity and the quantity ratio, the matching degree corresponding to the current data segment is obtained.

8. The neonatal sleep apnea early warning method according to any one of claims 1 to 4, characterized in that, The process of obtaining sleep risk values ​​based on various matching degrees includes: Based on each matching degree, a first matching degree and a preset number of second matching degrees are obtained; the timing of each second matching degree and the first matching degree is sequentially adjacent, and the first matching degree is the matching degree corresponding to the last data segment in each timing data segment; Fit each second matching degree to the first matching degree to obtain a fitted straight line; Based on the fitted line, the slope of the line is obtained; the slope of the line is the slope of the fitted line. The sleep risk value is obtained based on the slope of the straight line and the first matching degree.

9. The neonatal sleep apnea early warning method according to claim 8, characterized in that, After obtaining the sleep risk value based on the slope of the straight line and the first matching degree, the method further includes: Based on each matching degree, a second variance, a third matching degree, and a fourth matching degree are obtained; the second variance is the variance of each matching degree; the third matching degree is the maximum value among all matching degrees; and the fourth matching degree is the minimum value among all matching degrees. Based on the second variance, the third matching degree, and the fourth matching degree, the correction coefficient is obtained; The sleep risk value is corrected based on the correction factor.

10. A neonatal sleep apnea early warning device, characterized in that, include: The reading module is used to acquire initial audio data and monitor audio data during newborn sleep. The initial audio data is audio data acquired when the newborn is in a stable and normal breathing state; The processing module is used to acquire a dominant frequency signal based on the initial audio data; the dominant frequency signal is the signal generated by the newborn's breathing frequency during sleep. Furthermore, based on the monitored audio data, multiple time-series data segments are acquired; Furthermore, based on the main frequency signal and each time-series data segment, a matching degree corresponding to each time-series data segment is obtained; the matching degree is at least used to characterize the degree of similarity between the main frequency signal and the corresponding time-series data segment. Furthermore, based on each matching degree, a sleep risk value is obtained; The sleep risk value is used to characterize at least the probability of the newborn experiencing apnea; The early warning module is used to issue an early warning if the sleep risk value is greater than or equal to a first preset value.

Images