Breathing condition detection method and device

Abstract

The invention relates to the technical field of breathing condition monitoring, and discloses a breathing condition detection method and device.The method comprises the steps that ballistocardiogram signals of a target object are collected through a non-contact sensor, and breathing signals in the ballistocardiogram signals are extracted; performing envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal; analyzing the envelope line, and determining a target breathing segment according to an energy change point in the envelope line; determining a first respiration detection result according to the target respiration segment; inputting the target respiration segment into a preset auto-encoder model, performing auto-encoding reconstruction on the target respiration segment, and determining a second respiration detection result according to a reconstruction result; and determining the breathing condition of the target object according to the first breathing detection result and the second breathing detection result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of respiratory condition monitoring, and in particular to a respiratory condition detection method and device. BACKGROUND

[0002] At present, the detection of sleep apnea syndrome usually relies on polysomnography (PSG), but it is complex to wear, has poor body feeling, is high in cost, and is inconvenient for long-term screening at home. In the related art, respiratory detection can be performed in combination with time domain and frequency domain signals. This method relies on peak searching in the time domain signal and comparison between peaks, but when the signal-to-noise ratio is low, the bed mat coupling is poor, or the user's chest fluctuation is not effectively transmitted to the sensor, the sensitivity will decrease and the false positive rate will be high once the peak searching is wrong, and the respiratory detection accuracy is low. In addition, most of these methods rely on the combination of heart beat peak amplitude changes and respiratory peak value loss for judgment, and lack of processing mechanism for individual differences and respiratory signal waveforms, which further affects the respiratory detection accuracy. SUMMARY

[0003] The present application aims to at least solve the technical problems of low respiratory detection accuracy, and complex wearing, poor body feeling, high cost, and inconvenience for long-term screening at home in the related art.

[0004] To solve the above technical problems, the present application provides a respiratory condition detection method, comprising:

[0005] Collecting a heart impact signal of a target object by a non-contact sensor, and extracting a respiratory signal in the heart impact signal;

[0006] Performing envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal;

[0007] Analyzing the envelope line, and determining a target respiratory segment according to an energy change point in the envelope line;

[0008] Determining a first respiratory detection result according to the target respiratory segment;

[0009] Inputting the target respiratory segment into a preset autoencoder model, performing autoencoding reconstruction on the target respiratory segment, and determining a second respiratory detection result according to the reconstruction result;

[0010] Determining a respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result.

[0011] In some embodiments, analyzing the envelope line, and determining a target respiratory segment according to an energy change point in the envelope line, comprises:

[0012] performing continuous wavelet transform on the envelope data based on a Morlet mother wavelet to obtain continuous time-frequency feature data;

[0013] analyzing the continuous time-frequency feature data to extract a main respiratory frequency band energy and a bandwidth information;

[0014] determining an energy proportion of a preset respiratory frequency band in each time period according to the main respiratory frequency band energy and the bandwidth information;

[0015] obtaining an energy curve by performing a first preset time length moving average on the envelope;

[0016] detecting a start and end point of energy drop by performing a sliding window and cumulative sum processing on the energy curve;

[0017] determining the target respiratory segment according to the start and end point and the energy proportion of the preset respiratory frequency band in the sliding window segment.

[0018] In some embodiments, determining the target respiratory segment according to the start and end point and the energy proportion of the preset respiratory frequency band in the sliding window segment comprises:

[0019] if a duration of the start and end point is greater than a second preset time length and the respiratory frequency band energy proportion in the sliding window segment decreases, determining a respiratory signal segment in the sliding window segment as a first target respiratory segment;

[0020] if the duration of the start and end point is less than or equal to the second preset time length or the respiratory frequency band energy proportion in the sliding window segment does not decrease, determining the respiratory signal segment in the sliding window segment as a second target respiratory segment.

[0021] In some embodiments, determining a first respiratory detection result according to the target respiratory segment comprises:

[0022] if the target respiratory segment is the first target respiratory segment, determining a respiratory signal drop confidence of the first target respiratory segment as a first respiratory signal drop confidence;

[0023] if the target respiratory segment is the second target respiratory segment, determining a respiratory signal drop confidence of the second target respiratory segment as a second respiratory signal drop confidence;

[0024] wherein the second respiratory signal drop confidence is less than the first respiratory signal drop confidence.

[0025] In some embodiments, inputting the target respiratory segment into a preset autoencoder model to perform autoencoding reconstruction on the target respiratory segment, and determining a second respiratory detection result according to a reconstruction result comprises:

[0026] inputting the first target respiratory segment into the autoencoder model, performing autoencoding reconstruction on the first target respiratory segment, and determining a mean square error of the first target respiratory segment;

[0027] determining a reconstruction confidence of the first target respiratory segment according to the mean square error.

[0028] In some embodiments, the method further comprises:

[0029] extracting a heart rate signal in the ballistocardiogram signal;

[0030] determining a target heart rate segment according to the heart rate signal;

[0031] performing Fourier transform on the target heart rate segment to obtain a target frequency domain signal of the target heart rate segment;

[0032] calculating a short-time response of a target subject's heart rate according to the target frequency domain signal to obtain a third respiratory detection result;

[0033] determining a respiratory condition of the target subject according to the first respiratory detection result, the second respiratory detection result, and the third respiratory detection result.

[0034] In some embodiments, the target heart rate segment comprises a first target heart rate segment and a second target heart rate segment, and the second target heart rate segment is a heart rate segment before and after the first target heart rate segment by a third preset time length.

[0035] The calculating a short-time response of a target subject's heart rate according to the target frequency domain signal to obtain a third respiratory detection result comprises:

[0036] if the first target frequency domain signal of the first target heart rate segment is smaller than two second target frequency domain signals of the second target heart rate segment, determining a heart rate auxiliary confidence as a first heart rate auxiliary confidence;

[0037] if the first target frequency domain signal of the first target heart rate segment is smaller than one second target frequency domain signal of the second target heart rate segment, determining a heart rate auxiliary confidence as a first heart rate auxiliary confidence;

[0038] wherein the second heart rate auxiliary confidence is smaller than the first heart rate auxiliary confidence.

[0039] In some embodiments, the extracting a respiratory signal and a heart rate signal from the ballistocardiogram signal comprises:

[0040] performing band-pass filtering of 0.05-0.6 hz on the ballistocardiogram signal to extract the respiratory signal;

[0041] The heart impact signal is subjected to 0.8-3 Hz band-pass Butterworth filtering to extract the heart rate signal.

[0042] In some embodiments, the non-contact sensor is a fiber-optic sensor.

[0043] Embodiments of the present application also provide a respiratory condition detection device, which comprises:

[0044] a non-contact sensor configured to collect a heart impact signal of a target object;

[0045] a controller comprising:

[0046] an extraction module configured to extract a respiratory signal from the heart impact signal;

[0047] an envelope transformation module configured to perform envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal;

[0048] an envelope analysis module configured to analyze the envelope line and determine a target respiratory segment according to energy change points in the envelope line;

[0049] a first determination module configured to determine a first respiratory detection result according to the target respiratory segment;

[0050] a second determination module configured to input the target respiratory segment into a preset auto-encoder model, perform auto-encoding reconstruction on the target respiratory segment, and determine a second respiratory detection result according to the reconstruction result;

[0051] a respiratory detection module configured to determine a respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result.

[0052] The embodiment of the present application provides a respiratory condition detection method and device, a heart impact signal of a target object is collected through a non-contact sensor, and a respiratory signal in the heart impact signal is extracted; envelope transformation is performed on the respiratory signal to obtain an envelope line of the respiratory signal; the envelope line is analyzed, and a target respiratory segment is determined according to an energy change point in the envelope line; a first respiratory detection result is determined according to the target respiratory segment; the target respiratory segment is input into a preset autoencoder model, autoencoding reconstruction is performed on the target respiratory segment, and a second respiratory detection result is determined according to a reconstruction result; and the respiratory condition of the target object is determined according to the first respiratory detection result and the second respiratory detection result. The target respiratory segment that may have an abnormal respiratory condition can be accurately identified through envelope transformation and envelope analysis, the first respiratory detection result is obtained, the target respiratory segment is autoencoding reconstructed through the autoencoder model according to individual differences, the second respiratory detection result is obtained, then the two respiratory detection results are fused to accurately detect the respiratory condition of the target object, and the accuracy of respiratory detection is improved. In addition, the non-contact sensor is convenient to detect, and does not need to be directly contacted with the body, so that the discomfort of the target object is reduced, the comfort of health monitoring of the target object is improved, and the cost is low, and long-term screening at home is facilitated. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0054] Figure 1 The first flow chart of the respiratory condition detection method of the embodiment of the present application;

[0055] Figure 2 The second flow chart of the respiratory condition detection method of the embodiment of the present application;

[0056] Figure 3 The schematic diagram of the heart impact signal extracted by the respiratory condition detection method of the embodiment of the present application;

[0057] Figure 4 The schematic diagram of the respiratory signal extracted from the heart impact signal;

[0058] Figure 5 The schematic diagram of the envelope line obtained by performing envelope transformation on the respiratory signal;

[0059] Figure 6 The schematic diagram of the heart rate signal extracted from the heart impact signal;

[0060] Figure 7 a diagram of a frequency domain signal obtained by Fourier transforming a heart rate signal;

[0061] Figure 8 a diagram of a structure of a respiratory condition detection device according to an embodiment of the present application. DETAILED DESCRIPTION

[0062] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0063] It is to be understood that various alterations and modifications can be made to the embodiments of the application herein. Therefore, the above description should not be taken as limiting, but merely as exemplification of the embodiments of the application. Those skilled in the art will envision other modifications within the scope and spirit of the application.

[0064] The accompanying drawings incorporated in and forming a part of the specification illustrate embodiments of the present application and, together with the general description of the application given above, and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0065] These and other characteristics of the present application will become apparent from the following description of the preferred forms thereof given with reference to the accompanying drawings.

[0066] It is also to be understood that even though a number of embodiments of the present application have been described herein, the application should not be construed as limited thereto since modifications will readily occur to those skilled in the art having the benefit of the teachings presented herein. Therefore, numerous other embodiments of the present application will be apparent to those skilled in the art.

[0067] The above and other aspects, features, and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0068] Specific embodiments of the present application are described hereinbelow with reference to the accompanying drawings; however, it is to be understood that the embodiments described are merely exemplary of the application, which can be embodied in various forms. Well-known and / or redundant functions and structures are not described in detail to avoid obscuring the present application unnecessarily. Therefore, specific structural and functional details disclosed herein are not to be interpreted in a limiting manner, but merely as basis for the claims and representative basis for teaching one skilled in the art to employ the present application in virtually any appropriate detailed structure.

[0069] The specification can use phrases such as "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," which can refer to one or more embodiments of the same or different embodiments of the application.

[0070] Embodiment One

[0071] Figure 1 and Figure 2 A flow chart of a respiratory condition detection method according to an embodiment of the present application is shown. As shown in Figure 1 and Figure 2 The present application provides a respiratory condition detection method, which comprises:

[0072] S101: Collect a ballistocardiogram of a target object by a non-contact sensor, and extract a respiratory signal in the ballistocardiogram.

[0073] The target object is a to-be-detected user who needs to be detected for respiratory condition. The non-contact sensor is an optical fiber sensor, which can be installed on the back of a mattress, a seat cushion or a back cushion, and has no direct contact with the body of the target object for signal collection. The optical fiber sensor can acquire a ballistocardiogram (BCG) by sensing the small body movement and chest fluctuation of the human body. The optical fiber sensor can collect the ballistocardiogram of the target object in real time and send it to a controller of a respiratory condition detection device. As shown in Figure 3 , the signal is an original signal A, Figure 3 , in which the horizontal axis (X-axis) represents the sampling points (number of sampling points), and the vertical axis (Y-axis) represents the optical power. The sampling time of the optical fiber sensor is preferably 30 seconds, and the sampling frequency is preferably 50 Hz. The number of sampling points refers to the number of samples obtained by sampling the ballistocardiogram in a period of time, which is determined according to the sampling time and the sampling frequency. The number of sampling points = sampling time x sampling frequency.

[0074] In this embodiment, the optical fiber sensor can sample under the condition of single-channel 50 Hz sampling, which is convenient, fast, comfortable and low in cost.

[0075] The controller can extract the respiratory signal from the ballistocardiogram by filtering or other methods. Optionally, the ballistocardiogram can be subjected to a band-pass filtering of 0.05-0.6 Hz to extract a respiratory signal B, Figure 4 , as shown in Figure 4 , in which the horizontal axis represents the sampling points, and the vertical axis represents the optical power. The respiratory signal B contains waveform characteristics, which is convenient for subsequent analysis.

[0076] The frequency of the respiratory signal is relatively low (for example, the respiratory frequency of an adult is usually lower than 0.5 Hz), and in this embodiment, the respiratory signal B can be effectively extracted by band-pass filtering in the frequency band of 0.05-0.6 Hz, and the interference of other low-frequency signals can be reduced.

[0077] S102: Perform envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal.

[0078] The controller extracts the respiratory signal B, performs envelope transformation on the respiratory signal B, and obtains envelope line data of the respiratory signal.

[0079] Preferably, in this step, the Hilbert envelope transformation is performed on the respiratory signal. The Hilbert envelope transformation delays the phase of all frequency components of the respiratory signal by 90 degrees, thereby constructing an analytic signal. Then, the envelope line of the respiratory signal is obtained by calculating the modulus of the analytic signal. The envelope line C obtained by the Hilbert envelope transformation is as shown in Figure 5 Figure 5 The horizontal axis represents the sampling point, and the vertical axis represents the amplitude. The envelope line C can reflect the amplitude (or energy) variation trend of the respiratory signal B over time, and further reflect the speed of respiration. When the waveform of the respiratory signal B is smooth and the quality is high, the envelope line C is almost the same as the respiratory signal B itself. However, when the waveform of the respiratory signal B is large and the quality is poor, the envelope line C can improve the respiratory waveform, facilitating subsequent analysis of the envelope line. In specific implementations, other envelope transformation methods can also be used, such as wavelet transformation.

[0080] S103: Analyzing the envelope line to determine the target respiratory segment according to the energy variation points in the envelope line.

[0081] The controller detects the target respiratory segment through the envelope line variation points of the respiratory signal B. Compared with the detection method of determining the target respiratory segment through a fixed threshold sequence, the detection accuracy of the start and end boundaries of the target respiratory segment can be improved, and a more accurate target respiratory segment can be obtained.

[0082] S104: Determining the first respiratory detection result according to the target respiratory segment.

[0083] After obtaining the target respiratory segment by analyzing the envelope line of the respiratory signal B, the first respiratory detection result can be obtained by analyzing the target respiratory segment.

[0084] S105: Inputting the target respiratory segment into a preset autoencoder model, performing autoencoding reconstruction on the target respiratory segment, and determining the second respiratory detection result according to the reconstruction result.

[0085] After obtaining the target respiratory segment by analyzing the envelope line of the respiratory signal B, the second respiratory detection result can be obtained by inputting the target respiratory segment into a preset autoencoder model according to the individual differences of the target object and performing autoencoding reconstruction on the target respiratory segment.

[0086] The autoencoder model is an unsupervised autoencoder anomaly detection model, which can accurately detect abnormal respiratory conditions in the target respiratory segment. This model is more robust to individual differences and low signal-to-noise ratio conditions.

[0087] ​S106: Determine the respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result.

[0088] After the first respiratory detection result is directly determined through the target respiratory segment, and the second respiratory detection result is obtained by reconstructing the target respiratory segment using the autoencoder model, the two respiratory detection results are weighted and fused to obtain the final respiratory detection result.

[0089] The respiratory condition detection method provided by the embodiment of the present application can collect the ballistocardiogram of the target object through the non-contact sensor, and extract the respiratory signal in the ballistocardiogram; perform envelope transformation on the respiratory signal to obtain the envelope line of the respiratory signal; analyze the envelope line to determine the target respiratory segment according to the energy change point in the envelope line; determine the first respiratory detection result according to the target respiratory segment; input the target respiratory segment into the preset autoencoder model to perform autoencoding reconstruction on the target respiratory segment, and determine the second respiratory detection result according to the reconstruction result; and determine the respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result. The target respiratory segment that may have an abnormal respiratory condition can be accurately identified through envelope transformation and envelope analysis to obtain the first respiratory detection result, and the target respiratory segment is reconstructed through the autoencoder model according to individual differences to obtain the second respiratory detection result. Then, the two respiratory detection results are fused to accurately detect the respiratory condition of the target object, thereby improving the accuracy of respiratory detection. In addition, the non-contact sensor is convenient to detect, and does not need to be directly in contact with the body, thereby reducing the discomfort of the target object, improving the comfort of health monitoring of the target object, and being low in cost and convenient for long-term screening at home.

[0090] The optical fiber sensor has the advantages of high sensitivity, anti-electromagnetic interference, small size, safety, and multifunctionality, and is widely used in industries, medical treatment, communication, and other fields. The respiratory condition detection device based on the optical fiber sensor has the characteristics of small size and easy placement, is convenient to use, has a wide range of applications, and can meet the needs of long-term monitoring at home.

[0091] In some embodiments, in step S103, the envelope line is analyzed to determine the target respiratory segment according to the energy change point in the envelope line, including:

[0092] S1031: Perform continuous wavelet transformation on the envelope line data with Morlet mother wavelet as the base wave to obtain continuous time-frequency feature data;

[0093] S1032: Analyze the continuous time-frequency feature data to extract respiratory main frequency band energy and bandwidth information;

[0094] S1033: Determine the energy proportion of the preset respiratory frequency band in each time period according to the respiratory main frequency band energy and bandwidth information;

[0095] S1034: Obtain an energy curve by performing a sliding average on the envelope line for a first preset time length;

[0096] S1035: Detect start and end points of energy drop on the energy curve by using a sliding window and a cumulative sum process;

[0097] S1036: Determine a target respiratory segment according to the start and end points and an energy proportion of a preset respiratory frequency band in the sliding window segment.

[0098] In this step, a continuous wavelet transform (CWT) can be performed on the envelope line C using Morlet mother wavelet as the base wavelet to obtain continuous time-frequency feature data. Then, analysis is performed on the time-frequency feature data to extract respiratory main frequency band energy and bandwidth information, and the proportion of energy of each time period preset respiratory frequency band in the total energy is calculated. For example, the preset respiratory frequency band can be 0.1-0.35 hz.

[0099] Morlet wavelet, as a complex wavelet, has good time-frequency localization characteristics, and is particularly suitable for analysis and processing of non-stationary signals. Continuous wavelet transform uses Morlet wavelet as the mother wavelet to perform continuous wavelet transform on the signal, and the time-frequency spectrum of the signal can be obtained to directly display the frequency components of the signal changing with time.

[0100] Further, a sliding average is performed on the envelope line C to form an energy curve. The step length of the sliding average is preferably 1 second. Then, a sliding window (sliding preset time window) and a cumulative sum (CUSUM) process are performed on the energy curve to detect start and end points of significant energy drop in the sliding window segment. The parameters of the sliding window are preferably a sliding window step length of 1 second and a window length of 12 seconds.

[0101] In this embodiment, the mean or median value in the sliding window can be calculated in real time by using the sliding window, and used as the current threshold to quickly respond to changes in data. The cumulative sum identifies potential change points by accumulating the change amount of data.

[0102] In some embodiments, in step S1036, determining the target respiratory segment according to the start and end points and the energy proportion of the preset respiratory frequency band in the sliding window segment includes:

[0103] S201: If the duration of the start and end points is greater than a second preset time length, and the respiratory frequency band energy proportion in the sliding window segment decreases, determine the respiratory signal segment in the sliding window segment as a first target respiratory segment;

[0104] S202: If the duration of the start and end points is less than or equal to the second preset time length, or the respiratory frequency band energy proportion in the sliding window segment does not decrease, determine the respiratory signal segment in the sliding window segment as a second target respiratory segment.

[0105] When the duration of the start and end point of the significant drop of the energy in the sliding window is greater than 10s, and the proportion of the respiratory frequency band energy in the sliding window is also reduced, the respiratory signal segment is recorded as a first target respiratory segment, and the first target respiratory segment is a segment of respiratory abnormality.

[0106] If the duration of the start and end point is less than or equal to 10s, or the proportion of the respiratory frequency band energy in the sliding window is not reduced, the respiratory signal segment is recorded as a second target respiratory segment, and the target respiratory segment is a segment of normal respiration.

[0107] In some embodiments, in step S104, the first respiratory detection result is determined according to the target respiratory segment, including:

[0108] S1041: If the target respiratory segment is the first target respiratory segment, the respiratory signal drop confidence of the first target respiratory segment is determined as the first respiratory signal drop confidence;

[0109] S1042: If the target respiratory segment is the second target respiratory segment, the respiratory signal drop confidence of the second target respiratory segment is determined as the second respiratory signal drop confidence;

[0110] Wherein, the second respiratory signal drop confidence is less than the first respiratory signal drop confidence.

[0111] After determining the segment of respiratory abnormality and the segment of normal respiration, different respiratory signal drop confidences are assigned to obtain a preliminary respiratory state detection result. Since the first target respiratory segment is a segment of respiratory abnormality, the first respiratory signal drop confidence can be set to 1; the second target respiratory segment is a segment of normal respiration, and the second respiratory signal drop confidence can be set to 0. The respiratory signal drop confidence adopts 1 and 0 for convenient calculation.

[0112] In specific implementation, the first respiratory signal drop confidence can be a higher confidence than 1, such as 95% or 99%, and the specific value of the first respiratory signal drop confidence is not specifically limited in the present application; similarly, the second respiratory signal drop confidence can be a lower confidence than 0, such as 1% or 5%.

[0113] In some embodiments, in step S105, the target respiratory segment is input into a preset autoencoder model, the target respiratory segment is reconstructed by autoencoding, and the second respiratory detection result is determined according to the reconstruction result, including:

[0114] S1051: The first target respiratory segment is input into the autoencoder model, the first target respiratory segment is reconstructed by autoencoding, and the mean square error of the first target respiratory segment is determined.

[0115] S1052: Determine the reconstruction confidence of the first target respiratory segment according to the mean square error.

[0116] After determining the first target respiratory segment that may have a respiratory abnormality according to the respiratory energy change and the energy change duration, the first target respiratory segment is input into the autoencoder model, and the convolutional autoencoder is used for autoencoding reconstruction. Then, the mean square error of the first target respiratory segment is calculated, and then the reconstruction error of the first target respiratory segment is obtained, and the reconstruction confidence is determined according to the reconstruction error.

[0117] In this embodiment, the mean square error (MSE) is used to calculate the reconstruction error, and the mean μ and the standard deviation σ are recorded. When determining the reconstruction confidence, the threshold is μ+3σ, that is, the reconstruction confidence is MSE / (μ+3σ)-1.

[0118] The autoencoder model can be pre-constructed by a convolutional neural network. Specifically, human lying data can be collected first, and a large amount of training data can be collected. The optical fiber sensor can be placed under the mattress, and the target object lies on the mattress to collect a large number of respiratory segments. Then, abnormal respiratory segments are filtered, normal respiratory segments are extracted, and a normal respiratory data set is formed to prepare a training set for subsequent unsupervised models.

[0119] Then, the normal respiratory data set is input into the 1D-CNN (1D-Convolutional Neural Network, 1D-Convolutional Neural Network) autoencoder for training. The architecture of the 1D-CNN autoencoder is shown in Table 1. The input is a 1D convolution (Conv1d), and the Flatten layer (flatten layer) "flattens" the input, that is, one-dimensionalizes the multi-dimensional input, and outputs to the Dense layer (full connection layer), that is, the Flatten layer is used to transition from the convolutional layer to the fully connected layer; the Dense layer outputs the final result through a one-dimensional transpose convolution (ConvTranspose1d). Through training, an autoencoding model can be obtained. The 1D-CNN is specially used for processing sequence data (such as time series).

[0120] Table 1 Architecture of 1D-CNN Autoencoder

[0121]

[0122] In some embodiments, as shown in FIG. 1, Figure 2 the method further includes:

[0123] S301: Extracting a heart rate signal in the ballistocardiogram signal;

[0124] S302: Determining a target heart rate segment according to the heart rate signal;

[0125] S303: Perform Fourier transform on the target heart rate segment to obtain a target frequency domain signal of the target heart rate segment;

[0126] S304: Calculate a short-time response of the target object heart rate according to the target frequency domain signal to obtain a third respiration detection result;

[0127] S305: Determine the respiration condition of the target object according to the first respiration detection result, the second respiration detection result and the third respiration detection result.

[0128] In this embodiment, the respiration condition can be determined according to the heart rate signal. The heart rate signal can be extracted from the ballistocardiogram through filtering or the like.

[0129] Optionally, in step S301, the heart rate signal in the ballistocardiogram is extracted, including:

[0130] The ballistocardiogram is subjected to 0.8-3 hz band-pass Butterworth filtering to extract the heart rate signal. The extracted heart rate signal D is as shown in FIG. 4. Figure 6 Figure 6 The abscissa in FIG. 4 is also a sampling point, and the ordinate represents the optical power value.

[0131] According to the prior knowledge of the heart rate, the ideal heart rate of the user should be 55-70 times per minute, and therefore, the frequency of the heart rate signal should be about 0.9-1.7 hz. In this embodiment, the heart rate signal can be effectively extracted through band-pass Butterworth filtering in the 0.8-3 hz frequency band. The Butterworth filtering has the best flatness in the passband, and can maximize the preservation of the original information of the heart rate signal and reduce signal distortion. In this embodiment, the band-pass filtering and the Butterworth filtering are combined, which can not only accurately extract the heart rate signal in the required frequency band, but also can preserve the detailed features of the heart rate signal as much as possible and reduce signal distortion.

[0132] After the heart rate signal is extracted, the target heart rate segment with large heart rate change can be detected, and then Fourier transform (FFT) is performed thereon to obtain the frequency domain signal after Fourier transform as shown in FIG. 5. Figure 7 Figure 7 In FIG. 5, the abscissa represents the frequency, and the frequency gradually increases from left to right, representing the distribution of different frequency components in the signal, and the ordinate represents the amplitude.

[0133] After the frequency domain characteristics of the heart rate signal are obtained through the Fourier transform, the frequency with the highest peak value can be converted into the heart rate, the short-time response of the target object heart rate is calculated, the third respiration detection result is obtained according to the short-time response result, and then the first respiration detection result and the second respiration detection result are weighted and fused to obtain a more accurate respiration detection result of the target object.

[0134] ​​In specific implementation, the short-time reaction of the target heart rate can be calculated by calculating the mean, slope, delay, etc. of the heart rate (HR). In some embodiments, the target heart rate segment includes a first target heart rate segment and a second target heart rate segment, and the second target heart rate segment is a heart rate segment before and after the first target heart rate segment by a third preset time length.

[0135] In step S304, the short-time reaction of the target heart rate is calculated according to the target frequency domain signal, and a third respiration detection result is obtained, including:

[0136] In S3041, if the first target frequency domain signal of the first target heart rate segment is smaller than the two second target frequency domain signals of the second target heart rate segment, the heart rate auxiliary confidence is determined as the first heart rate auxiliary confidence.

[0137] In S3042, if the first target frequency domain signal of the first target heart rate segment is smaller than one second target frequency domain signal of the second target heart rate segment, the heart rate auxiliary confidence is determined as the second heart rate auxiliary confidence.

[0138] In S3042, if the first target frequency domain signal of the first target heart rate segment is smaller than one second target frequency domain signal of the second target heart rate segment, the heart rate auxiliary confidence is determined as the second heart rate auxiliary confidence.

[0139] In this embodiment, the target heart rate segment extracted from the heart rate signal includes not only the segment (the first target heart rate segment) that may have heart rate abnormalities, but also the heart rate segments (the second target heart rate segment) in the adjacent time period, which are used as a comparison standard to accurately detect the heart rate abnormal segment.

[0140] When the heart rate of the first target heart rate segment that may have heart rate abnormalities is smaller than the heart rates of the two second target heart rate segments adjacent thereto, it is determined that the first target heart rate segment has a greater impact on the respiration signal, and the heart rate auxiliary confidence can be set as the first heart rate auxiliary confidence. Since the heart rate of the first target heart rate segment is smaller than the heart rates of the two second target heart rate segments adjacent thereto, the first heart rate auxiliary confidence can be set larger, for example, 1. When the heart rate of the first target heart rate segment that may have heart rate abnormalities is smaller than the heart rate of one of the two second target heart rate segments adjacent thereto, it is determined that part of the heart rate signal has an impact on the respiration signal, and the heart rate auxiliary confidence can be set as the second heart rate auxiliary confidence. Since the heart rate of the first target heart rate segment is smaller than the heart rate of only one of the two second target heart rate segments adjacent thereto, the second heart rate auxiliary confidence can be set smaller, for example, 0.5.

[0141] In specific implementation, the specific values of the first heart rate auxiliary confidence and the second heart rate auxiliary confidence can be determined according to actual conditions, and the second heart rate auxiliary confidence is half of the first heart rate auxiliary confidence.

[0142] After obtaining the third respiration detection result, the respiration detection judgment confidence of the weighted fusion of the first respiration detection result, the second respiration detection result and the third respiration detection result can be expressed as:

[0143] Respiration signal confidence = ω1 x respiration signal drop confidence + ω2 x reconstruction confidence + ω3 x heart rate auxiliary confidence.

[0144] As can be seen from the above, the final respiration detection result is obtained by weighting and fusing the respiration detection results according to the weights of different influencing factors. Since the first respiration detection result of the respiration signal itself is the most important for respiration detection, the weight ω1 of the first respiration detection result is the largest. The second respiration detection result is obtained by using the encoder model to perform self-encoding reconstruction on the target respiration segment, which is also a direct analysis of the respiration segment, so the weight ω2 of the second respiration detection result is relatively large. The third respiration detection result is obtained by analyzing the heart rate to assist in judging the respiration detection result, so the weight ω3 of the third respiration detection result is the smallest. For example, in the present embodiment, ω1 = 0.5, ω2 = 0.4, ω3 = 0.1, and ω1 + ω2 + ω3 = 1. The respiration state of the target object is determined according to the respiration signal confidence. If the final respiration signal confidence > 0.75, it is determined that the target object has apnea, and “apnea” is output. If the final respiration signal confidence < 0.5, it is determined that the target object does not have apnea, and “no apnea” is output. If the final respiration signal confidence is between 0.5 and 0.75, it is determined that the respiration state of the target object is uncertain, and “undetectable / need to review” is output.

[0145] In the present embodiment, the fusion of the respiration signal and the short-time reaction of the pre-and post-heart rate reduces the false positive rate of respiration detection.

[0146] In summary, the respiration condition detection method provided in the present embodiment combines the time domain signal features of the respiration signal with the pre-and post-heart rate changes of the frequency domain signal features of the heart rate signal, and combines the self-encoder model to weight and detect the respiration state of the target object, which can stably and effectively identify the occurrence of apnea. At the same time, the present application supports short window (for example, 30 seconds) segment analysis, and can output the transition state of “undetectable / need to review” under boundary conditions, which is particularly suitable for home long-term monitoring and meets the daily respiration monitoring of the target object.

[0147] Embodiment Two

[0148] Figure 8 The overall block diagram of the respiration condition detection device of the present embodiment is shown. As shown in Figure 8 The present embodiment also provides a respiration condition detection device, which comprises:

[0149] The non-contact sensor 10 is configured to collect a ballistocardiogram of a target object;

[0150] The controller 20 comprises: an extraction module configured to extract a respiratory signal in the ballistocardiogram; an envelope transformation module configured to perform envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal; an envelope analysis module configured to analyze the envelope line, and determine a target respiratory segment according to an energy change point in the envelope line; a first determination module configured to determine a first respiratory detection result according to the target respiratory segment; a second determination module configured to input the target respiratory segment into a preset autoencoder model, perform autoencoding reconstruction on the target respiratory segment, and determine a second respiratory detection result according to a reconstruction result; and a respiratory detection module configured to determine a respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result.

[0151] In some embodiments, the envelope analysis module is further configured to:

[0152] Perform continuous wavelet transformation on the envelope line data with Morlet mother wavelet as a base wave to obtain continuous time-frequency feature data;

[0153] Perform analysis on the continuous time-frequency feature data to extract respiratory main frequency band energy and bandwidth information;

[0154] Determine an energy proportion of a preset respiratory frequency band in each time period according to the respiratory main frequency band energy and the bandwidth information;

[0155] Obtain an energy curve by first preset length sliding average on the envelope line;

[0156] Detect start and end points of energy drop by sliding window and cumulative sum processing on the energy curve;

[0157] Determine the target respiratory segment according to the start and end points and the energy proportion of the preset respiratory frequency band in the sliding window segment.

[0158] In some embodiments, the envelope analysis module is further configured to:

[0159] If a duration of the start and end points is greater than a second preset length, and the respiratory frequency band energy proportion in the sliding window segment decreases, determine a respiratory signal segment in the sliding window segment as a first target respiratory segment;

[0160] If the duration of the start and end points is less than or equal to the second preset length, or the respiratory frequency band energy proportion in the sliding window segment does not decrease, determine the respiratory signal segment in the sliding window segment as a second target respiratory segment.

[0161] In some embodiments, the first determination module is further configured to:

[0162] If the target respiratory segment is the first target respiratory segment, the controller 20 is further configured to determine the respiratory signal decrease confidence of the first target respiratory segment as a first respiratory signal decrease confidence;

[0163] If the target respiratory segment is the second target respiratory segment, the controller 20 is further configured to determine the respiratory signal decrease confidence of the second target respiratory segment as a second respiratory signal decrease confidence;

[0164] The second respiratory signal decrease confidence is less than the first respiratory signal decrease confidence.

[0165] In some embodiments, the second determining module is further configured to:

[0166] input the first target respiratory segment into the autoencoder model to perform autoencoding reconstruction on the first target respiratory segment, and determine a mean square error of the first target respiratory segment;

[0167] determine a reconstruction confidence of the first target respiratory segment according to the mean square error.

[0168] In some embodiments, the extracting module is further configured to extract a heart rate signal in the ballistocardiogram;

[0169] The controller 20 further includes a third determining module configured to:

[0170] determine a target heart rate segment according to the heart rate signal;

[0171] perform Fourier transform on the target heart rate segment to obtain a target frequency domain signal of the target heart rate segment;

[0172] calculate a short-term response of the target object heart rate according to the target frequency domain signal to obtain a third respiratory detection result;

[0173] The respiratory detection module is further configured to determine a respiratory condition of the target object according to the first respiratory detection result, the second respiratory detection result, and the third respiratory detection result.

[0174] In some embodiments, the target heart rate segment includes a first target heart rate segment and a second target heart rate segment, and the second target heart rate segment is a heart rate segment before and after the first target heart rate segment by a third preset time length; the controller 20 is further configured to:

[0175] If the first target frequency domain signal of the first target heart rate segment decreases compared to two second target frequency domain signals of the second target heart rate segment, the controller 20 is further configured to determine the heart rate auxiliary confidence as a first heart rate auxiliary confidence;

[0176] If the first target frequency domain signal of the first target heart rate segment decreases compared to one second target frequency domain signal of the second target heart rate segment, the controller 20 is further configured to determine the heart rate auxiliary confidence as a second heart rate auxiliary confidence;

[0177] The second heart rate auxiliary confidence is less than the first heart rate auxiliary confidence.

[0178] In some embodiments, the extraction module is further configured to:

[0179] The heart impact signal is subjected to 0.05-0.6 hz band-pass filtering to extract a respiratory signal.

[0180] The heart impact signal is subjected to 0.8 hz-3 hz band-pass Butterworth filtering to extract a heart rate signal.

[0181] In some embodiments, the non-contact sensor is a fiber-optic sensor.

[0182] The respiratory condition detection device provided by the embodiments of the present application corresponds to the respiratory condition detection method of the above-mentioned embodiments, and any optional items in the respiratory condition detection method embodiments are also applicable to the embodiments of the respiratory condition detection device, which will not be described here.

[0183] Embodiment three

[0184] The embodiments of the present application also provide an electronic device comprising at least a memory and a processor, the memory having a computer program stored thereon, and the processor implementing the above-mentioned respiratory condition detection method when executing the computer program stored on the memory.

[0185] In some embodiments, the processor executing the computer program can be a processing device comprising one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), and the like. More specifically, the processor can be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor operating other instruction sets, or a processor operating a combination of instruction sets. The processor can also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), and the like.

[0186] The memory can be a read-only memory (ROM), a random access memory (RAM), a phase change random access memory (PRAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), an electrically erasable programmable read-only memory (EEPROM), other types of random access memory (RAM), a flash disk or other forms of flash memory, a cache, a register, a static memory, a compact disc read-only memory (CD-ROM), a digital versatile disc (DVD) or other optical memory, a magnetic tape or other magnetic storage device, or any other possible non-transitory medium used to store information or instructions accessible by a computer device, and the like.

[0187] The electronic device of the embodiments of the present application can include, but is not limited to, fixed terminal devices such as servers, desktop computers, digital TVs, and the like, and mobile terminal devices such as in-vehicle devices (for example, head-up display devices), handheld devices (for example, mobile phones, tablet computers, and the like), wearable devices (for example, smart watches, smart bands, and the like), and the like.

[0188] Embodiment Four

[0189] The embodiments of the present application also provide a computer readable storage medium, the computer readable medium stores a computer program, and the computer program is executed by a processor to implement the above-mentioned respiratory condition detection method.

[0190] The computer readable storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. In the embodiments of the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus, for example, the above-mentioned memory.

[0191] The computer program of the embodiments of the present application can be organized into one or more computer executable components or modules. Any number and combination of such components or modules can be used to implement aspects of the present application. For example, aspects of the present application are not limited to the specific computer executable instructions or specific components or modules illustrated in the figures and described herein. Other embodiments can include different computer executable instructions or components that have more or less functionality than that described herein.

[0192] The above description is merely preferred embodiments of the present application and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the disclosure involved in the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features disclosed in the present application (but not limited to) having similar functions to form technical solutions.

Claims

1. A respiratory condition detection method, characterized by, The method comprises the following steps: Collecting a heart impact signal of a target object by a non-contact sensor, and extracting a respiratory signal in the heart impact signal; Performing envelope transformation on the respiratory signal to obtain an envelope line of the respiratory signal; Analyzing the envelope line to determine a target respiratory segment according to energy change points in the envelope line; Determining a first respiratory detection result according to the target respiratory segment; Inputting the target respiratory segment into a preset autoencoder model to perform autoencoding reconstruction on the target respiratory segment, and determining a second respiratory detection result according to the reconstruction result; Determining a respiratory condition of the target object according to the first respiratory detection result and the second respiratory detection result.

2. The method of claim 1, wherein, The method comprises the following steps: Performing continuous wavelet transformation on the envelope line data with Morlet mother wavelet as a basic wave to obtain continuous time-frequency feature data; Analyzing the continuous time-frequency feature data to extract respiratory main frequency band energy and bandwidth information; Determining energy proportion of a preset respiratory frequency band in each time period according to the respiratory main frequency band energy and bandwidth information; Obtaining an energy curve by performing first preset time length sliding average on the envelope line; Detecting start and end points of energy drop by performing sliding window and cumulative sum processing on the energy curve; Determining the target respiratory segment according to the start and end points and energy proportion of the preset respiratory frequency band in the sliding window segment.

3. The method of claim 2, wherein, The method comprises the following steps: If the duration of the start and end points is greater than a second preset time length and the respiratory frequency band energy proportion in the sliding window segment decreases, determining that a respiratory signal segment in the sliding window segment is a first target respiratory segment; If the duration of the start and end points is less than or equal to the second preset time length or the respiratory frequency band energy proportion in the sliding window segment does not decrease, determining that a respiratory signal segment in the sliding window segment is a second target respiratory segment.

4. The method of claim 3, wherein, The method comprises the following steps: If the target respiratory segment is the first target respiratory segment, determining that a respiratory signal drop confidence of the first target respiratory segment is a first respiratory signal drop confidence; If the target respiratory segment is the second target respiratory segment, determining that a respiratory signal drop confidence of the second target respiratory segment is a second respiratory signal drop confidence; The second respiratory signal drop confidence is less than the first respiratory signal drop confidence.

5. The method of claim 3, wherein, The method comprises the following steps: Inputting the first target respiratory segment into the autoencoder model to perform autoencoding reconstruction on the first target respiratory segment, determining a mean square error of the first target respiratory segment; Determining a reconstruction confidence of the first target respiratory segment according to the mean square error.

6. The method of claim 1, wherein, The method further comprises the following steps: Extracting a heart rate signal in the heart impact signal; Determining a target heart rate segment according to the heart rate signal; performing Fourier transform on the target heart rate segment to obtain a target frequency domain signal of the target heart rate segment; calculating a short-time response of a target object heart rate according to the target frequency domain signal to obtain a third respiration detection result; determining a respiration condition of the target object according to the first respiration detection result, the second respiration detection result and the third respiration detection result.

7. The method of claim 6, wherein, The target heart rate segment includes a first target heart rate segment and a second target heart rate segment, and the second target heart rate segment is a heart rate segment before and after the first target heart rate segment by a third preset time length. The method for calculating a short-time response of a target object heart rate according to the target frequency domain signal to obtain a third respiration detection result includes: If the first target frequency domain signal of the first target heart rate segment is smaller than the two second target frequency domain signals of the second target heart rate segment, the heart rate auxiliary confidence is determined as a first heart rate auxiliary confidence; If the first target frequency domain signal of the first target heart rate segment is smaller than one second target frequency domain signal of the second target heart rate segment, the heart rate auxiliary confidence is determined as a second heart rate auxiliary confidence; The second heart rate auxiliary confidence is smaller than the first heart rate auxiliary confidence.

8. The method of claim 1, wherein, The method for extracting a respiration signal and a heart rate signal from the ballistocardiogram includes: performing 0.05-0.6 hz band-pass filtering on the ballistocardiogram to extract the respiration signal; performing 0.8 hz-3 hz band-pass Butterworth filtering on the ballistocardiogram to extract the heart rate signal.

9. The method according to any one of claims 1 to 8, characterized in that, The non-contact sensor is an optical fiber sensor.

10. A respiratory condition detection apparatus, characterized by, The method includes: a non-contact sensor configured to collect a ballistocardiogram of a target object; a controller including: an extraction module configured to extract a respiration signal from the ballistocardiogram; an envelope transformation module configured to perform envelope transformation on the respiration signal to obtain an envelope line of the respiration signal; an envelope analysis module configured to analyze the envelope line to determine a target respiration segment according to energy change points in the envelope line; a first determination module configured to determine a first respiration detection result according to the target respiration segment; a second determination module configured to input the target respiration segment into a preset autoencoder model to perform autoencoding reconstruction on the target respiration segment, and determine a second respiration detection result according to a reconstruction result; a respiration detection module configured to determine a respiration condition of the target object according to the first respiration detection result and the second respiration detection result.

Images