A wearable sensing system and its respiration wave generation and respiration rate estimation method

Abstract

The application discloses a wearable sensing system and a respiration wave generation and respiration rate estimation method thereof, relates to the field of respiration processing, and solves the problem that respiration wave and respiration rate cannot be accurately analyzed based on electrocardiogram, and comprises a data processing module, a stacked convolution module, a respiration discrimination module, a signal demodulation module and a data calculation module; the data processing module is used for initial processing of electrocardiogram of a person to be detected; the stacked convolution module is used for stacked convolution of voltage intensity sequences of the heart of the person to be detected and convolution kernels in different convolution layers to obtain respiration wave sequences of the person to be detected; the respiration discrimination module is used for discriminating the respiration wave sequences of the person to be detected; the signal demodulation module is used for signal demodulation of the respiration waves of the person to be detected to obtain final respiration waves; and the data calculation module is used for calculating the respiration rate of the person to be detected, and the application realizes accurate analysis of respiration wave and respiration rate based on electrocardiogram.

Description

Technical Field

[0001] This invention belongs to the field of respiratory data processing technology, specifically a wearable sensing system and its respiratory wave generation and respiratory rate estimation method. Background Technology

[0002] The respiratory wave is a physical signal representation of respiratory physiological movements. Its generation depends on the sensing and detection of respiratory-related physical quantities. The respiratory rate refers to the number of breaths per unit time (usually expressed in "breaths / minute"). Its calculation is based on the generated respiratory wave signal, identifying the respiratory cycle and counting the frequency.

[0003] In the existing technology, because respiratory waves, respiratory rate and heart rate are strongly related, there are errors in the methods for generating respiratory waves and estimating respiratory rate, and wearable devices cannot be effectively used to analyze the patient's heart rate. Therefore, it is not possible to further realize the generation of the patient's respiratory waves and the prediction of the patient's respiratory rate based on electrocardiogram.

[0004] Therefore, this invention proposes a wearable sensing system and a method for generating respiratory waves and estimating respiratory rate. Summary of the Invention

[0005] The purpose of this invention is to propose a wearable sensing system and a method for generating respiratory waves and estimating respiratory rate, so as to solve the problem mentioned in the background art that it is impossible to accurately analyze respiratory waves and respiratory rate based on electrocardiogram.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] In a first aspect, a wearable sensing system includes a data acquisition module, a data processing module, a stacked convolution module, a breathing discrimination module, a signal demodulation module, and a data calculation module;

[0008] The data acquisition module is used to acquire the electrocardiogram (ECG) of the person to be tested at a fixed acquisition frequency, and send the ECG of the person to be tested to the data processing module; the data processing module is used to perform initial processing on the ECG of the person to be tested, to obtain the voltage intensity sequence and sequence length of the heart of the person to be tested and send them to the stacked convolution module, to obtain the continuous acquisition duration of the ECG and send it to the respiration discrimination module;

[0009] The stacked convolution module is used to stack and convolve the voltage intensity sequence of the heart of the person under test with convolution kernels in different convolutional layers. The stacked convolution yields the respiratory wave sequence of the person under test, which is then sent to the respiratory discrimination module. The respiratory discrimination module is used to discriminate the respiratory wave sequence of the person under test, generate a demodulated signal, and send it to the signal demodulation module. The signal demodulation module is used to demodulate the respiratory wave of the person under test after receiving the demodulated signal, obtain the final respiratory wave, and send it to the data calculation module. The data calculation module is used to calculate the respiratory rate of the person under test after receiving the final respiratory wave.

[0010] Furthermore, the data processing module's processing procedure includes the following:

[0011] Obtain the start and end times of the electrocardiogram (ECG) acquisition, subtract the start time from the end time to get the continuous acquisition duration of the ECG, and multiply the continuous acquisition duration by a fixed acquisition frequency to get the number of time points in the ECG.

[0012] The voltage intensity of the heart of the person under test at different time points is obtained, and the voltage intensity is sorted in ascending order according to time to obtain the voltage intensity sequence of the heart of the person under test.

[0013] The number of time points in the electrocardiogram is used as the sequence length of the voltage intensity sequence.

[0014] Furthermore, the stacked convolution process of the stacked convolution module includes the following:

[0015] The voltage intensity sequence is denoted as [A1, A2, ..., As], where s is the time node number, which is a positive integer, and As is the voltage intensity of the human heart to be tested at the s-th time node.

[0016] The convolutional layers are respectively labeled as pre-convolutional layer, first convolutional layer, second convolutional layer, ..., b-th convolutional layer and output convolutional layer; where b is the number of the convolutional layer and b is a positive integer;

[0017] Different dilation factors are assigned to different convolutional layers. The dilation factor for the pre-convolutional layer and the output convolution is 0, and the dilation factor for the first convolutional layer is 2. 0 The expansion factor of the second convolutional layer is 2. 1 The expansion factor of the third convolutional layer is 2. 2 And so on, the inflation factor of the b-th convolutional layer is 2. b-1 ;

[0018] Obtain the convolution kernel and dilation factor corresponding to the first convolutional layer, and then (2 0 -1) Insert zeros between adjacent elements in the convolution kernel, and denote the corresponding convolution kernel as the first dilated convolution kernel;

[0019] Similarly, obtain the convolution kernel and dilation factor corresponding to the b-th convolutional layer, and then (2 b-1 -1) zeros are inserted between adjacent elements in the convolution kernel, and the corresponding convolution kernel is denoted as the b-th dilated convolution kernel;

[0020] Obtain the initial length CSb of the convolutional kernel corresponding to different convolutional layers, and then use the formula CDb=(CSb-1)×2 b-1 +1 calculates the kernel length CDb of the dilated convolution kernel;

[0021] The length of the first dilated convolution kernel is denoted as the first convolution kernel length, and the length of the second dilated convolution kernel is denoted as the second convolution kernel length; similarly, the length of the b-th dilated convolution kernel is denoted as the b-th convolution kernel length.

[0022] Furthermore, the stacked convolution process of the stacked convolution module also includes the following:

[0023] Obtain the corresponding convolution kernel in the preconvolution layer, and convolve the convolution kernel in the preconvolution layer with the voltage intensity sequence to obtain the first sequence;

[0024] Subtract one from the length of the first convolution kernel and divide by two to get the first padding quantity of the first sequence. Obtain the first element and the last element in the first sequence, and pad the first element with the first padding quantity of zeros and the last element with the first padding quantity of zeros. Record the first sequence after padding as the first padding sequence.

[0025] The first padding sequence is convolved with the first dilated convolution kernel to obtain the second sequence.

[0026] Subtract one from the length of the second convolution kernel and divide by two to get the second padding number of the second sequence. Obtain the first and last elements in the second sequence, and pad the first element with the second padding number of zeros and the last element with the second padding number of zeros. Record the padded second sequence as the second padding sequence.

[0027] The second padding sequence is convolved with the second dilated convolution kernel to obtain the third sequence.

[0028] Similarly, the third sequence is padded to obtain the third padded sequence, and the third padded sequence is convolved with the third dilated convolution kernel to obtain the fourth sequence. This process is repeated, and the b-th sequence is padded to obtain the b-th padded sequence, and the b-th padded sequence is convolved with the b-th dilated convolution kernel to obtain the b+1-th sequence.

[0029] The elements at the same position in the first sequence, the second sequence, the third sequence, ..., the b+1th sequence are summed to obtain the stacked convolution sequence;

[0030] The stacked convolutional sequence is replaced using the ReLU activation function. The replacement process is as follows:

[0031] Each element in the stacked convolutional sequence is compared with zero;

[0032] If all elements in the stacked convolution sequence are greater than or equal to zero, no operation is performed;

[0033] If any element in the stacked convolution sequence is less than zero, then the corresponding element is replaced with zero;

[0034] The replaced stacked convolution sequence is convolved with the convolution kernel corresponding to the output convolution layer to obtain the respiratory wave sequence of the person to be tested.

[0035] Furthermore, the discrimination process of the breathing discrimination module is as follows:

[0036] Construct a Cartesian coordinate system with time as the horizontal axis and respiratory amplitude as the vertical axis, and draw respiratory feature points in the Cartesian coordinate system. Connect different respiratory feature points to form a respiratory wave. Record the respiratory feature points whose respiratory amplitude is greater than the respiratory amplitudes of the left and right adjacent sides as the maximum feature points.

[0037] The number of maximal feature points in the Cartesian coordinate system is counted, the continuous acquisition time of the electrocardiogram is obtained, and the number of maximal feature points is divided by the continuous acquisition time and then multiplied by a fixed time to obtain the number of breaths of the person being tested within the fixed time.

[0038] The number of breaths of the person to be tested within a fixed time period is compared with the standard respiratory rate range;

[0039] If the number of breaths of the person being tested does not fall within the standard respiratory rate range within a fixed time period, the electrocardiogram of the person being tested is repeatedly stacked and convolved until the number of breaths of the person being tested falls within the standard respiratory rate range within a fixed time period.

[0040] If the number of breaths of the person being tested within a fixed time period falls within the standard respiratory rate range, a demodulation signal is generated.

[0041] Furthermore, the demodulation process of the signal demodulation module includes the following:

[0042] Obtain the impulse response of the bandpass filter, and convolve the respiratory wave with the impulse response of the bandpass filter to obtain the filtered respiratory wave. The respiratory wave is composed of multiple sets of different superimposed signals, and the expression of the corresponding filtered respiratory wave is denoted as:

[0043] Where x(t) is the respiratory wave, k=1,2,...,i, k is the number of the superimposed signal, u k (t) represents the k-th superimposed signal, where t is time;

[0044] The Hilbert transform of different superimposed signals yields the one-sided spectrum of the corresponding superimposed signal, where the one-sided spectrum of the superimposed signal is: [δ(t) + j / (πt)]*u k (t), where δ(t) is the Dirac function, j is the imaginary unit, π is pi, and the symbol "*" is the convolution symbol;

[0045] Through formula The fundamental frequency band JP of different superimposed signals was calculated. k ,in w is the frequency shift factor in the frequency domain. k Let k be the center frequency of the superimposed signal;

[0046] The base bandwidth JD of different superimposed signals is calculated using the formula. k The formula is as follows:

[0047] Where ∂t is a partial differential operator, square L 2 Norm;

[0048] Will As a constraint, the base bandwidths of different superimposed signals are summed to obtain the total bandwidth ZDK of different superimposed signals. The minimum value of the total bandwidth is obtained by iterating and comparing different bandwidth sums, and the corresponding superimposed signal is denoted as the intrinsic mode function.

[0049] Furthermore, the demodulation process of the signal demodulation module also includes the following:

[0050] Performing an N-point Discrete Fourier Transform on the respiratory wave x(t) yields the Discrete Fourier Transform signals X(f) of the respiratory wave at different real-time frequencies. The squared magnitude of the Discrete Fourier Transform signal |X(f)| is then calculated using the formula. 2 The formula is as follows:

[0051] |X(f)| 2 =Re[X(f)] 2 +Im[X(f)] 2 Where Re[] is the symbol for taking the real part and Im[] is the symbol for taking the imaginary part;

[0052] Obtain the start and end times of the filtered respiratory wave, and subtract the start time from the end time to obtain the duration T of the respiratory wave.

[0053] The formula GLP(f) = |X(f)| is used. 2 / T calculates the power spectral density GLP(f) of the respiratory wave.

[0054] Similarly, N-point discrete Fourier transforms are performed on different intrinsic mode functions, and the power spectral density of different intrinsic mode functions is calculated.

[0055] A Cartesian coordinate system is constructed with real-time frequency as the horizontal axis and power spectral density as the vertical axis. Based on the power spectral density corresponding to the fixed frequency range, the power spectral feature points of the respiratory wave are plotted in the Cartesian coordinate system. The power spectral feature points of different power spectral densities are connected to obtain the power spectral density map of the respiratory wave.

[0056] Similarly, power spectral density plots for different intrinsic mode functions are obtained;

[0057] The power spectral density of different power spectral feature points in the power spectral density map corresponding to the respiratory wave is traversed and compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the original peak frequency of the respiratory wave.

[0058] For any intrinsic mode function, the power spectral density of different power spectral feature points in the power spectral density plot corresponding to the intrinsic mode function is compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the actual peak frequency of the corresponding intrinsic mode function.

[0059] Similarly, the actual peak frequencies of different intrinsic mode functions are obtained;

[0060] Multiple peak absolute error values ​​are obtained by subtracting the original peak frequency from different actual peak frequencies and taking the absolute value. The minimum peak absolute error value is obtained by iterating and comparing different peak absolute error values, and the intrinsic mode function corresponding to the minimum peak absolute error value is recorded as the final respiratory wave.

[0061] Furthermore, the calculation process of the data calculation module includes the following:

[0062] Constructing an autoregressive model:

[0063] Where y(n) is the final respiratory wave at time node n, y(nm) is the final respiratory wave at time node nm, and a m Here, e(n) is the weight coefficient, p is the error term, and n and m are both time nodes.

[0064] Applying the Z-transform to both sides of the autoregressive model yields:

[0065] Where Y(Z) is the Z-transform of y(n), E(Z) is the Z-transform of e(n), and Z is a complex number;

[0066] Combining like terms yields:

[0067] ;

[0068] The system function H(Z) of the autoregressive model is calculated using the following formula:

[0069] ;

[0070] Where z1, z2, ..., z p For constant terms;

[0071] Obtain the passband cutoff frequency of the bandpass filter, and multiply the passband cutoff frequency by a fixed value to obtain the first and second respiratory counts;

[0072] Compare the first number of breaths with the second number of breaths;

[0073] If the first respiratory count is greater than the second respiratory count, then the respiratory rate interval is constructed with the second respiratory count as the left endpoint and the first respiratory count as the right endpoint;

[0074] If the first respiratory count is less than the second respiratory count, then the respiratory rate interval is constructed with the first respiratory count as the left endpoint and the second respiratory count as the right endpoint.

[0075] The number of respiratory rates within a respiratory rate interval is obtained by subtracting the left endpoint from the right endpoint of the interval. The respiratory rate interval is then divided into different segments at equal intervals based on the number of respiratory rates, with the endpoint of each segment being the point to be analyzed.

[0076] Furthermore, the calculation process of the data calculation module also includes the following:

[0077] The frequency response corresponding to different points to be analyzed within the respiratory rate interval is calculated. The specific calculation process is as follows:

[0078] For any point to be analyzed within the respiratory rate range, obtain the respiratory rate corresponding to the point to be analyzed, and divide the respiratory rate by a fixed value to obtain the actual frequency corresponding to the point to be analyzed.

[0079] Divide the actual frequency by the fixed acquisition frequency and multiply by 2π to obtain the digital angular frequency θ of the point to be analyzed. Replace Z in the system function with... Obtain the frequency response of the corresponding point to be analyzed. ;

[0080] Through formula The magnitude response of the point to be analyzed is calculated. ;

[0081] Similarly, the amplitude response at different points to be analyzed within the respiratory rate range was calculated;

[0082] The maximum value of the amplitude response is obtained by iterating through and comparing the amplitude responses of different analysis points, and the respiratory rate of the analysis point corresponding to the maximum value of the amplitude response is recorded as the estimated respiratory rate.

[0083] Secondly, a method for generating respiratory waves and estimating respiratory rate in a wearable sensing system, the method comprising the following steps:

[0084] Step S101: Process the electrocardiogram of the person to be tested to obtain the voltage intensity sequence, sequence length, and continuous acquisition duration of the electrocardiogram of the person to be tested.

[0085] Step S102: Stack the voltage intensity sequence with convolution kernels in different convolutional layers to obtain the respiratory wave sequence of the person to be tested.

[0086] Step S103: The respiratory wave sequence obtained by stacked convolution is judged to determine whether the obtained respiratory wave sequence is accurate;

[0087] Step S104: Demodulate the respiratory wave of the person to be tested to obtain the final respiratory wave of the person to be tested;

[0088] Step S105: Calculate the respiratory rate of the person being tested based on the final respiratory wave.

[0089] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0090] 1. The present invention first processes the electrocardiogram of the person to be tested to obtain the voltage intensity sequence, sequence length and continuous acquisition duration of the electrocardiogram of the person to be tested. Then, the voltage intensity sequence is stacked and convolved with the convolution kernels in different convolution layers to obtain the respiratory wave sequence of the person to be tested. The present invention can accurately extract the respiratory wave of the person to be tested from the electrocardiogram.

[0091] 2. The present invention also discriminates the respiratory wave sequence obtained by stacked convolution, thereby determining whether the respiratory wave sequence obtained by stacked convolution is accurate. At the same time, when the respiratory wave sequence is accurate, the respiratory wave of the person to be tested is demodulated to obtain the final respiratory wave of the person to be tested. Finally, the respiratory rate of the person to be tested is calculated based on the final respiratory wave. The present invention realizes accurate calculation of the respiratory rate of the person to be tested based on electrocardiogram. Attached Figure Description

[0092] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0093] Figure 1 This is an overall system block diagram of the present invention;

[0094] Figure 2 This is a schematic diagram of stacked convolution in this invention;

[0095] Figure 3 This is a schematic diagram of the convolution of the pre-convolution layer in this invention;

[0096] Figure 4 This is a schematic diagram of the respiratory wave in this invention;

[0097] Figure 5 This is the power spectral density diagram of the respiratory wave in this invention;

[0098] Figure 6 This is a frequency response diagram of different points to be analyzed in this invention;

[0099] Figure 7 This is a flowchart of the method in this invention. Detailed Implementation

[0100] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0101] Example 1, please refer to Figure 1 - Figure 5 As shown, the technical solution provided by the present invention is: a wearable sensing system, the system including a data acquisition module, a data processing module, a stacked convolution module, a respiration discrimination module, a signal demodulation module, and a data calculation module. The system first inputs the voltage intensity of the heart of the person to be tested at different time points into the generation model, and stacks and convolves the voltage intensity with the convolution kernels in different convolution layers in the generation model to generate the respiratory wave of the person to be tested. Then, the generated respiratory wave is discriminated to determine whether the generated respiratory wave is normal. Next, the generated respiratory wave is filtered and demodulated to obtain the demodulated final respiratory wave. Finally, the amplitude response of the final respiratory wave is calculated to calculate the estimated respiratory rate of the person to be tested.

[0102] In this embodiment, the data acquisition module is used to acquire the electrocardiogram of the person to be tested at a fixed acquisition frequency, and send the electrocardiogram of the person to be tested to the data processing module;

[0103] In practice, wearable sensing devices are used to test the personnel being tested and obtain the corresponding electrocardiograms;

[0104] Specifically, the data processing module is used to perform initial processing on the electrocardiogram of the person to be tested. The processing procedure is as follows:

[0105] Obtain the start and end times of the electrocardiogram (ECG) acquisition, subtract the start time from the end time to get the continuous acquisition duration of the ECG, and multiply the continuous acquisition duration by a fixed acquisition frequency to get the number of time points in the ECG.

[0106] The voltage intensity of the heart of the person under test at different time points is obtained, and the voltage intensity is sorted in ascending order according to time to obtain the voltage intensity sequence of the heart of the person under test.

[0107] The number of time points in the electrocardiogram is used as the sequence length of the voltage intensity sequence;

[0108] It should be explained that the electrocardiogram (ECG) of the person being tested is actually an analysis of the voltage intensity of the person's heart at different time points. The horizontal axis of the ECG represents time, and the vertical axis represents voltage intensity.

[0109] The data processing module sends the voltage intensity sequence and sequence length of the heart of the person to be tested to the stacked convolution module, and sends the continuous acquisition duration of the electrocardiogram to the respiration discrimination module.

[0110] In the specific implementation process, the stacked convolution module is used to stack the voltage intensity sequence of the heart of the person to be tested with convolution kernels in different convolutional layers. The stacked convolution process is as follows:

[0111] The voltage intensity sequence is denoted as [A1, A2, ..., As], where s is the time node number, which is a positive integer, and As is the voltage intensity of the human heart to be tested at the s-th time node.

[0112] Please see Figure 2 As shown, the convolutional layers are respectively labeled as pre-convolutional layer, first convolutional layer, second convolutional layer, ..., b-th convolutional layer and output convolutional layer; where b is the number of the convolutional layer and b is a positive integer;

[0113] Different dilation factors are assigned to different convolutional layers. The dilation factor for the pre-convolutional layer and the output convolution is 0, and the dilation factor for the first convolutional layer is 2. 0 The expansion factor of the second convolutional layer is 2. 1 The expansion factor of the third convolutional layer is 2. 2 And so on, the inflation factor of the b-th convolutional layer is 2. b-1 ;

[0114] Obtain the convolution kernel and dilation factor corresponding to the first convolutional layer, and then (2 0-1) Insert zeros between adjacent elements in the convolution kernel, and denote the corresponding convolution kernel as the first dilated convolution kernel;

[0115] In practice, the convolution kernel is a sequence or matrix composed of weight coefficients. In this embodiment, the convolution kernel exists in the form of a sequence.

[0116] Similarly, obtain the convolution kernel and dilation factor corresponding to the b-th convolutional layer, and then (2 b-1 -1) zeros are inserted between adjacent elements in the convolution kernel, and the corresponding convolution kernel is denoted as the b-th dilated convolution kernel;

[0117] Obtain the initial length CSb of the convolutional kernel corresponding to different convolutional layers, and then use the formula CDb=(CSb-1)×2 b-1 +1 calculates the kernel length CDb of the dilated convolution kernel;

[0118] In this embodiment, the length of the convolution kernels corresponding to the pre-convolutional layer and the output convolutional layer is 1, and the length of the convolution kernels corresponding to the remaining convolutional layers is equal before dilation. It should be explained that the dilation factor represents the insertion of different numbers of zeros between adjacent elements of the convolution kernel, thereby increasing the size of the convolution kernel. For example, when b=2 and the convolution kernel is [1, 1, 1], the kernel length is 3, and the dilation factor is 2. Inserting 2-1 zeros between adjacent elements of the convolution kernel results in an dilated convolution kernel [1, 0, 1, 0, 1], and the corresponding length of the dilated convolution kernel is CD=(3-1)×2. 2-1 +1=5;

[0119] The length of the first dilated convolution kernel is denoted as the first convolution kernel length, and the length of the second dilated convolution kernel is denoted as the second convolution kernel length; similarly, the length of the b-th dilated convolution kernel is denoted as the b-th convolution kernel length.

[0120] Please see Figure 3 As shown, the corresponding convolution kernel in the preconvolution layer is obtained, and the convolution kernel in the preconvolution layer is convolved with the voltage intensity sequence to obtain the first sequence;

[0121] In practice, convolution is an integral transformation operation that generates a third function from two functions. Essentially, it's the integral of the product of the overlapping portions of the functions after flipping and translating. For an example, please refer to [link to example]. Figure 3As shown, the first sequence is obtained by convolving the convolution kernel with the voltage intensity sequence. First, the first element in the convolution kernel is aligned with the first element in the voltage intensity sequence. Then, the values ​​at the same position are multiplied to obtain M, 2N, and 3J. The three values ​​are added together to obtain the value corresponding to the first position in the first sequence. Then, the convolution kernel is moved to the right by one length, and the values ​​at the same position are multiplied again and then added together to obtain the value corresponding to the second position in the first sequence, 2M+3N+4J. Finally, the convolution kernel is moved to the right by one length again, and the values ​​are multiplied and added together to obtain the value corresponding to the last position in the first sequence.

[0122] Subtract one from the length of the first convolution kernel and divide by two to get the first padding quantity of the first sequence. Obtain the first element and the last element in the first sequence, and pad the first element with the first padding quantity of zeros and the last element with the first padding quantity of zeros. Record the first sequence after padding as the first padding sequence.

[0123] For example, if the first dilated convolution kernel is [1, 0, 1, 0, 1], the length of the first convolution kernel is 5, and the first sequence is [1, 1, 1], then the first padding quantity is equal to 2, and the first padding sequence is [0, 0, 1, 1, 1, 0, 0].

[0124] The first padding sequence is convolved with the first dilated convolution kernel to obtain the second sequence.

[0125] Specifically, the convolution process of the first padding sequence and the first dilated convolution kernel is the same as the convolution process described above, and will not be repeated here;

[0126] Subtract one from the length of the second convolution kernel and divide by two to get the second padding number of the second sequence. Obtain the first and last elements in the second sequence, and pad the first element with the second padding number of zeros and the last element with the second padding number of zeros. Record the padded second sequence as the second padding sequence.

[0127] The second padding sequence is convolved with the second dilated convolution kernel to obtain the third sequence.

[0128] Similarly, the third sequence is padded to obtain the third padded sequence, and the third padded sequence is convolved with the third dilated convolution kernel to obtain the fourth sequence. This process is repeated, and the b-th sequence is padded to obtain the b-th padded sequence, and the b-th padded sequence is convolved with the b-th dilated convolution kernel to obtain the b+1-th sequence.

[0129] The elements at the same position in the first sequence, the second sequence, the third sequence, ..., the b+1th sequence are summed to obtain the stacked convolution sequence;

[0130] The stacked convolutional sequence is replaced using the ReLU activation function. The replacement process is as follows:

[0131] Each element in the stacked convolutional sequence is compared with zero;

[0132] If all elements in the stacked convolution sequence are greater than or equal to zero, no operation is performed;

[0133] If any element in the stacked convolution sequence is less than zero, then the corresponding element is replaced with zero;

[0134] The replaced stacked convolution sequence is convolved with the convolution kernel corresponding to the output convolution layer to obtain the respiratory wave sequence of the person to be tested;

[0135] It should be noted that the time point corresponding to each element in the respiratory wave sequence is the same as the time point corresponding to each element in the voltage intensity sequence, and each element in the respiratory wave sequence represents the respiratory amplitude of the person being tested.

[0136] The stacked convolution module sends the respiratory wave sequence of the person to be tested to the respiratory discrimination module.

[0137] In this embodiment, the respiratory discrimination module is used to discriminate the respiratory wave sequence of the person to be tested. The discrimination process is as follows:

[0138] Construct a Cartesian coordinate system with time as the horizontal axis and respiratory amplitude as the vertical axis, and draw respiratory feature points in the Cartesian coordinate system. Connect different respiratory feature points to form a respiratory wave. Record the respiratory feature points whose respiratory amplitude is greater than the respiratory amplitudes of the left and right adjacent sides as the maximum feature points.

[0139] For example, please refer to Figure 4 As shown, the horizontal axis of the coordinate system represents time, and the vertical axis represents amplitude. The original signal is the respiratory wave.

[0140] The number of maximal feature points in the Cartesian coordinate system is counted, the continuous acquisition time of the electrocardiogram is obtained, and the number of maximal feature points is divided by the continuous acquisition time and then multiplied by a fixed time to obtain the number of breaths of the person being tested within the fixed time.

[0141] In this embodiment, the fixed duration is 60 seconds, the maximum feature point is the transition point from inhalation to exhalation of the person being tested, and the number of maximum feature points is counted as the number of breaths during the continuous collection period;

[0142] The number of breaths of the person to be tested within a fixed time period is compared with the standard respiratory rate range;

[0143] If the number of breaths of the person being tested does not fall within the standard respiratory rate range within a fixed time period, the electrocardiogram of the person being tested is repeatedly stacked and convolved until the number of breaths of the person being tested falls within the standard respiratory rate range within a fixed time period.

[0144] If the number of breaths of the person being tested within a fixed time period falls within the standard respiratory rate range, a demodulation signal is generated;

[0145] The breathing discrimination module sends the demodulated signal to the signal demodulation module.

[0146] In this embodiment, the signal demodulation module is used to demodulate the respiratory wave of the person to be tested after receiving the demodulated signal. The demodulation process is as follows:

[0147] Obtain the impulse response of the bandpass filter, and convolve the respiratory wave with the impulse response of the bandpass filter to obtain the filtered respiratory wave. The respiratory wave is composed of multiple sets of different superimposed signals, and the expression of the corresponding filtered respiratory wave is denoted as:

[0148] Where x(t) is the respiratory wave, k=1,2,...,i, k is the number of the superimposed signal, u k (t) represents the k-th superimposed signal, where t is time;

[0149] In this embodiment, the passband cutoff frequencies of the bandpass filter are 0.1Hz and 0.5Hz, respectively, and the impulse response of the bandpass filter is... i equals 5, indicating that the respiratory wave is composed of 5 different superimposed signals;

[0150] It should be explained that the respiratory wave is composed of multiple sets of superimposed signals. For example, the respiratory wave may be composed of five superimposed signals A, B, C, D and E, or it may be composed of five superimposed signals V, W, X, Y and Z.

[0151] The Hilbert transform of different superimposed signals yields the one-sided spectrum of the corresponding superimposed signal, where the one-sided spectrum of the superimposed signal is: [δ(t) + j / (πt)]*u k (t), where δ(t) is the Dirac function, j is the imaginary unit, π is pi, and the symbol "*" is the convolution symbol;

[0152] Through formula The fundamental frequency band JP of different superimposed signals was calculated. k ,in w is the frequency shift factor in the frequency domain. k Let k be the center frequency of the superimposed signal;

[0153] The base bandwidth JD of different superimposed signals is calculated using the formula.k The formula is as follows:

[0154] Where ∂t is a partial differential operator, square L 2 Norm;

[0155] Will As a constraint, the base bandwidths of different superimposed signals are summed to obtain the total bandwidth ZDK of different superimposed signals. The minimum value of the total bandwidth is obtained by iterating and comparing different bandwidth sums, and the corresponding superimposed signal is recorded as the intrinsic mode function.

[0156] It should be explained that the constraint condition means that when the base bandwidths of different superimposed signals are added together, the superimposed signals must be from the same group of superimposed signals, that is, the superimposed signals can be restored to the respiratory wave signal after being superimposed.

[0157] Performing an N-point Discrete Fourier Transform on the respiratory wave x(t) yields the Discrete Fourier Transform signals X(f) of the respiratory wave at different real-time frequencies. The squared magnitude of the Discrete Fourier Transform signal |X(f)| is then calculated using the formula. 2 The formula is as follows:

[0158] |X(f)| 2 =Re[X(f)] 2 +Im[X(f)] 2 Where Re[] is the symbol for taking the real part and Im[] is the symbol for taking the imaginary part;

[0159] For example, if X(f) = 1 + 2j, then Re[X(f)] = 1 and Im[X(f)] = 2;

[0160] Obtain the start and end times of the filtered respiratory wave, and subtract the start time from the end time to obtain the duration T of the respiratory wave.

[0161] The formula GLP(f) = |X(f)| is used. 2 / T calculates the power spectral density GLP(f) of the respiratory wave.

[0162] Similarly, N-point discrete Fourier transforms are performed on different intrinsic mode functions, and the power spectral density of different intrinsic mode functions is calculated.

[0163] A Cartesian coordinate system is constructed with real-time frequency as the horizontal axis and power spectral density as the vertical axis. Based on the power spectral density corresponding to the fixed frequency range, the power spectral feature points of the respiratory wave are plotted in the Cartesian coordinate system. The power spectral feature points of different power spectral densities are connected to obtain the power spectral density map of the respiratory wave.

[0164] Similarly, power spectral density plots for different intrinsic mode functions are obtained;

[0165] In this embodiment, the fixed frequency range is [0, 1] Hz. For an example, please refer to [link to example]. Figure 5 As shown, the power spectral density of the original signal is the same as the power spectral density of the respiratory wave.

[0166] The power spectral density of different power spectral feature points in the power spectral density map corresponding to the respiratory wave is traversed and compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the original peak frequency of the respiratory wave.

[0167] For any intrinsic mode function, the power spectral density of different power spectral feature points in the power spectral density plot corresponding to the intrinsic mode function is compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the actual peak frequency of the corresponding intrinsic mode function.

[0168] Similarly, the actual peak frequencies of different intrinsic mode functions are obtained;

[0169] Multiple peak absolute error values ​​are obtained by subtracting the original peak frequency from different actual peak frequencies and taking the absolute value. The minimum peak absolute error value is obtained by iterating and comparing different peak absolute error values. The intrinsic mode function corresponding to the minimum peak absolute error value is recorded as the final respiratory wave.

[0170] For example, the original peak frequency of the respiratory wave signal is 0.25 Hz. There are 5 intrinsic mode functions. The actual peak frequencies corresponding to the first to the fifth intrinsic mode functions are 0.12 Hz, 1 Hz, 0.81 Hz, 0.5 Hz and 0.26 Hz, respectively. Subtracting the original peak frequency from the different actual peak frequencies and taking the absolute value, the peak absolute error values ​​are 0.13 Hz, 0.75 Hz, 0.56 Hz, 0.25 Hz and 0.01 Hz, respectively. Among them, 0.01 Hz is the minimum value of the peak absolute error value. The fifth intrinsic mode function is then recorded as the final respiratory wave.

[0171] The signal demodulation module sends the final respiratory wave to the data calculation module.

[0172] In this embodiment, the data calculation module is used to calculate the respiratory rate of the person being tested after receiving the final respiratory wave. The calculation process is as follows:

[0173] Constructing an autoregressive model:

[0174] Where y(n) is the final respiratory wave at time node n, y(nm) is the final respiratory wave at time node nm, and a mHere, e(n) is the weight coefficient, p is the error term, and n and m are both time nodes.

[0175] In this embodiment, the model order p equals 8, and the weight parameter a can be estimated using the existing Burg algorithm. m With error term e(n);

[0176] Applying the Z-transform to both sides of the autoregressive model yields:

[0177] Where Y(Z) is the Z-transform of y(n), E(Z) is the Z-transform of e(n), and Z is a complex number;

[0178] Combining like terms yields:

[0179] ;

[0180] The system function H(Z) of the autoregressive model is calculated using the following formula:

[0181] ;

[0182] Where z1, z2, ..., z p For constant terms;

[0183] It needs to be explained that... After raising both sides to the power of Z, the numerator becomes Z to the power of p, while the denominator becomes a polynomial with the highest order Z to the power of p. Factoring the corresponding polynomial in the numerator yields the final form of the system function.

[0184] Obtain the passband cutoff frequency of the bandpass filter, and multiply the passband cutoff frequency by a fixed value to obtain the first and second respiratory counts;

[0185] Compare the first number of breaths with the second number of breaths;

[0186] If the first respiratory count is greater than the second respiratory count, then the respiratory rate interval is constructed with the second respiratory count as the left endpoint and the first respiratory count as the right endpoint;

[0187] If the first respiratory count is less than the second respiratory count, then the respiratory rate interval is constructed with the first respiratory count as the left endpoint and the second respiratory count as the right endpoint.

[0188] Subtract the left endpoint of the respiratory rate interval from the right endpoint of the respiratory rate interval to obtain the number of respiratory rates within the interval. Divide the respiratory rate interval into different segments at equal intervals based on the number of respiratory rates, where the endpoint of each segment is the point to be analyzed.

[0189] For example, the passband cutoff frequencies of the bandpass filter are 0.1Hz and 0.5Hz, and the fixed value is 60. 0.1×60=6bpm and 0.5×60=30bpm, so the respiratory rate range is [6, 30]bpm. Subtracting 6 from 30 gives 24, so the respiratory rate range is divided into 24 equally spaced segments.

[0190] The frequency response corresponding to different points to be analyzed within the respiratory rate interval is calculated. The specific calculation process is as follows:

[0191] For any point to be analyzed within the respiratory rate range, obtain the respiratory rate corresponding to the point to be analyzed, and divide the respiratory rate by a fixed value to obtain the actual frequency corresponding to the point to be analyzed.

[0192] Divide the actual frequency by the fixed acquisition frequency and multiply by 2π to obtain the digital angular frequency θ of the point to be analyzed. Replace Z in the system function with... Obtain the frequency response of the corresponding point to be analyzed. ;

[0193] Through formula The magnitude response of the point to be analyzed is calculated. ;

[0194] Similarly, the amplitude response at different points to be analyzed within the respiratory rate range was calculated;

[0195] The maximum value of the amplitude response is obtained by iterating and comparing the amplitude responses of different points to be analyzed, and the respiratory rate corresponding to the maximum value of the amplitude response at the point to be analyzed is recorded as the estimated respiratory rate.

[0196] For clarification, the amplitude response refers to the amplitude of the corresponding frequency response; for examples, please refer to [link to example]. Figure 6 As shown, a coordinate system is constructed with respiratory rate on the horizontal axis and respiratory amplitude on the vertical axis. Figure 6 The respiratory rate corresponding to the position in the middle circle is the estimated respiratory rate.

[0197] Example 2, please refer to Figure 7 As shown, based on another concept of the same invention, a method for generating respiratory waves and estimating respiratory rate in a wearable sensing system is proposed, the method comprising the following steps:

[0198] Step S101: Process the electrocardiogram of the person to be tested to obtain the voltage intensity sequence, sequence length, and continuous acquisition duration of the electrocardiogram of the person to be tested.

[0199] Step S102: Stack the voltage intensity sequence with convolution kernels in different convolutional layers to obtain the respiratory wave sequence of the person to be tested.

[0200] Step S103: The respiratory wave sequence obtained by stacked convolution is judged to determine whether the obtained respiratory wave sequence is accurate;

[0201] Step S104: Demodulate the respiratory wave of the person to be tested to obtain the final respiratory wave of the person to be tested;

[0202] Step S105: Calculate the respiratory rate of the person being tested based on the final respiratory wave.

[0203] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A wearable sensing system, characterized in that, It includes a data acquisition module, a data processing module, a stacked convolution module, a breathing discrimination module, a signal demodulation module, and a data calculation module; The data acquisition module is used to acquire the electrocardiogram (ECG) of the person to be tested at a fixed acquisition frequency, and send the ECG of the person to be tested to the data processing module; the data processing module is used to perform initial processing on the ECG of the person to be tested, to obtain the voltage intensity sequence and sequence length of the heart of the person to be tested and send them to the stacked convolution module, to obtain the continuous acquisition duration of the ECG and send it to the respiration discrimination module; The stacked convolution module is used to stack the voltage intensity sequence of the heart of the person to be tested with the convolution kernels in different convolution layers. The stacked convolution obtains the respiratory wave sequence of the person to be tested and sends it to the respiratory discrimination module. The respiratory discrimination module is used to discriminate the respiratory wave sequence of the person to be tested, generate a demodulated signal, and send it to the signal demodulation module; the signal demodulation module is used to demodulate the respiratory wave of the person to be tested after receiving the demodulated signal, obtain the final respiratory wave, and send it to the data calculation module; the data calculation module is used to calculate the respiratory rate of the person to be tested after receiving the final respiratory wave. The stacked convolution process of the stacked convolution module includes the following: The voltage intensity sequence is denoted as [A1, A2, ..., As], where s is the time node number, which is a positive integer, and As is the voltage intensity of the human heart to be tested at the s-th time node. The convolutional layers are respectively labeled as pre-convolutional layer, first convolutional layer, second convolutional layer, ..., b-th convolutional layer and output convolutional layer; where b is the number of the convolutional layer and b is a positive integer; Different dilation factors are assigned to different convolutional layers. The dilation factor for the pre-convolutional layer and the output convolution is 0, and the dilation factor for the first convolutional layer is 2. 0 The expansion factor of the second convolutional layer is 2. 1 The expansion factor of the third convolutional layer is 2. 2 And so on, the inflation factor of the b-th convolutional layer is 2. b-1 ; Obtain the convolution kernel and dilation factor corresponding to the first convolutional layer, and then (2 0 -1) zeros are inserted between adjacent elements in the convolution kernel, and the corresponding convolution kernel is denoted as the first dilated convolution kernel; Similarly, obtain the convolution kernel and dilation factor corresponding to the b-th convolutional layer, and then (2 b-1 -1) zeros are inserted between adjacent elements in the convolution kernel, and the corresponding convolution kernel is denoted as the b-th dilated convolution kernel; Obtain the initial length CSb of the convolutional kernel corresponding to different convolutional layers, and then use the formula CDb=(CSb-1)×2 b-1 +1 calculates the kernel length CDb of the dilated convolution kernel; The length of the first dilated convolution kernel is denoted as the first convolution kernel length, and the length of the second dilated convolution kernel is denoted as the second convolution kernel length; similarly, the length of the b-th dilated convolution kernel is denoted as the b-th convolution kernel length.

2. The wearable sensing system according to claim 1, characterized in that, The data processing module's processing procedure includes the following: Obtain the start and end times of the electrocardiogram (ECG) acquisition, subtract the start time from the end time to get the continuous acquisition duration of the ECG, and multiply the continuous acquisition duration by a fixed acquisition frequency to get the number of time points in the ECG. The voltage intensity of the heart of the person under test at different time points is obtained, and the voltage intensity is sorted in ascending order according to time to obtain the voltage intensity sequence of the heart of the person under test. The number of time points in the electrocardiogram is used as the sequence length of the voltage intensity sequence.

3. The wearable sensing system according to claim 1, characterized in that, The stacked convolution process of the stacked convolution module also includes the following: Obtain the corresponding convolution kernel in the preconvolution layer, and convolve the convolution kernel in the preconvolution layer with the voltage intensity sequence to obtain the first sequence; Subtract one from the length of the first convolution kernel and divide by two to get the first padding quantity of the first sequence. Obtain the first element and the last element in the first sequence, and pad the first element with the first padding quantity of zeros and the last element with the first padding quantity of zeros. Record the first sequence after padding as the first padding sequence. The first padding sequence is convolved with the first dilated convolution kernel to obtain the second sequence. Subtract one from the length of the second convolution kernel and divide by two to get the second padding number of the second sequence. Obtain the first and last elements in the second sequence, and pad the first element with the second padding number of zeros and the last element with the second padding number of zeros. Record the padded second sequence as the second padding sequence. The second padding sequence is convolved with the second dilated convolution kernel to obtain the third sequence. Similarly, the third sequence is padded to obtain the third padded sequence, and the third padded sequence is convolved with the third dilated convolution kernel to obtain the fourth sequence. This process is repeated, and the b-th sequence is padded to obtain the b-th padded sequence, and the b-th padded sequence is convolved with the b-th dilated convolution kernel to obtain the b+1-th sequence. The elements at the same position in the first sequence, the second sequence, the third sequence, ..., the b+1th sequence are summed to obtain the stacked convolution sequence; The stacked convolutional sequence is replaced using the ReLU activation function. The replacement process is as follows: Each element in the stacked convolutional sequence is compared with zero; If all elements in the stacked convolution sequence are greater than or equal to zero, no operation is performed; If any element in the stacked convolution sequence is less than zero, then the corresponding element is replaced with zero; The replaced stacked convolution sequence is convolved with the convolution kernel corresponding to the output convolution layer to obtain the respiratory wave sequence of the person to be tested.

4. The wearable sensing system according to claim 1, characterized in that, The specific discrimination process of the respiratory discrimination module is as follows: Construct a Cartesian coordinate system with time as the horizontal axis and respiratory amplitude as the vertical axis, and draw respiratory feature points in the Cartesian coordinate system. Connect different respiratory feature points to form a respiratory wave. Record the respiratory feature points whose respiratory amplitude is greater than the respiratory amplitudes of the left and right adjacent sides as the maximum feature points. The number of maximal feature points in the Cartesian coordinate system is counted, the continuous acquisition time of the electrocardiogram is obtained, and the number of maximal feature points is divided by the continuous acquisition time and then multiplied by a fixed time to obtain the number of breaths of the person being tested within the fixed time. The number of breaths of the person to be tested within a fixed time period is compared with the standard respiratory rate range; If the number of breaths of the person being tested does not fall within the standard respiratory rate range within a fixed time period, the electrocardiogram of the person being tested is repeatedly stacked and convolved until the number of breaths of the person being tested falls within the standard respiratory rate range within a fixed time period. If the number of breaths of the person being tested within a fixed time period falls within the standard respiratory rate range, a demodulation signal is generated.

5. A wearable sensing system according to claim 1, characterized in that, The demodulation process of the signal demodulation module includes the following: Obtain the impulse response of the bandpass filter, and convolve the respiratory wave with the impulse response of the bandpass filter to obtain the filtered respiratory wave. The respiratory wave is composed of multiple sets of different superimposed signals, and the expression of the corresponding filtered respiratory wave is denoted as: Where x(t) is the respiratory wave, k=1,2,...,i, k is the number of the superimposed signal, u k (t) represents the k-th superimposed signal, where t is time; The Hilbert transform of different superimposed signals yields the one-sided spectrum of the corresponding superimposed signal, where the one-sided spectrum of the superimposed signal is: [δ(t) + j / (πt)]*u k (t), where δ(t) is the Dirac function, j is the imaginary unit, π is pi, and the symbol "*" is the convolution symbol; Through formula The fundamental frequency band JP of different superimposed signals was calculated. k ,in w is the frequency shift factor in the frequency domain. k Let k be the center frequency of the superimposed signal; The base bandwidth JD of different superimposed signals is calculated using the formula. k The formula is as follows: Where ∂t is a partial differential operator, square L 2 Norm; Will As a constraint, the base bandwidths of different superimposed signals are summed to obtain the total bandwidth ZDK of different superimposed signals. The minimum value of the total bandwidth is obtained by iterating and comparing different bandwidth sums, and the corresponding superimposed signal is denoted as the intrinsic mode function.

6. A wearable sensing system according to claim 5, characterized in that, The demodulation process of the signal demodulation module also includes the following: Performing an N-point Discrete Fourier Transform on the respiratory wave x(t) yields the Discrete Fourier Transform signals X(f) of the respiratory wave at different real-time frequencies. The squared magnitude of the Discrete Fourier Transform signal |X(f)| is then calculated using the formula. 2 The formula is as follows: |X(f)| 2 =Re[X(f)] 2 +Im[X(f)] 2 Where Re[] is the symbol for taking the real part and Im[] is the symbol for taking the imaginary part; Obtain the start and end times of the filtered respiratory wave, and subtract the start time from the end time to obtain the duration T of the respiratory wave. The formula GLP(f) = |X(f)| is used. 2 / T calculates the power spectral density GLP(f) of the respiratory wave. Similarly, N-point discrete Fourier transforms are performed on different intrinsic mode functions, and the power spectral density of different intrinsic mode functions is calculated. A Cartesian coordinate system is constructed with real-time frequency as the horizontal axis and power spectral density as the vertical axis. Based on the power spectral density corresponding to the fixed frequency range, the power spectral feature points of the respiratory wave are plotted in the Cartesian coordinate system. The power spectral feature points of different power spectral densities are connected to obtain the power spectral density map of the respiratory wave. Similarly, power spectral density plots for different intrinsic mode functions are obtained; The power spectral density of different power spectral feature points in the power spectral density map corresponding to the respiratory wave is traversed and compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the original peak frequency of the respiratory wave. For any intrinsic mode function, the power spectral density of different power spectral feature points in the power spectral density plot corresponding to the intrinsic mode function is compared to obtain the maximum value of the power spectral density, and the real-time frequency corresponding to the maximum value of the power spectral density is recorded as the actual peak frequency of the corresponding intrinsic mode function. Similarly, the actual peak frequencies of different intrinsic mode functions are obtained; Multiple peak absolute error values ​​are obtained by subtracting the original peak frequency from different actual peak frequencies and taking the absolute value. The minimum peak absolute error value is obtained by iterating and comparing different peak absolute error values, and the intrinsic mode function corresponding to the minimum peak absolute error value is recorded as the final respiratory wave.

7. A wearable sensing system according to claim 1, characterized in that, The calculation process of the data calculation module includes the following: Constructing an autoregressive model: Where y(n) is the final respiratory wave at time node n, y(nm) is the final respiratory wave at time node nm, and a m Here, e(n) is the weight coefficient, p is the error term, and n and m are both time nodes. Applying the Z-transform to both sides of the autoregressive model yields: Where Y(Z) is the Z-transform of y(n), E(Z) is the Z-transform of e(n), and Z is a complex number; Combining like terms yields: ； The system function H(Z) of the autoregressive model is calculated using the following formula: ； Where z1, z2, ..., z p For constant terms; Obtain the passband cutoff frequency of the bandpass filter, and multiply the passband cutoff frequency by a fixed value to obtain the first and second respiratory counts; Compare the first number of breaths with the second number of breaths; If the first respiratory count is greater than the second respiratory count, then the respiratory rate interval is constructed with the second respiratory count as the left endpoint and the first respiratory count as the right endpoint; If the first respiratory count is less than the second respiratory count, then the respiratory rate interval is constructed with the first respiratory count as the left endpoint and the second respiratory count as the right endpoint. The number of respiratory rates within a respiratory rate interval is obtained by subtracting the left endpoint from the right endpoint of the interval. The respiratory rate interval is then divided into different segments at equal intervals based on the number of respiratory rates, with the endpoint of each segment being the point to be analyzed.

8. A wearable sensing system according to claim 7, characterized in that, The calculation process of the data calculation module also includes the following: The frequency response corresponding to different points to be analyzed within the respiratory rate interval is calculated. The specific calculation process is as follows: For any point to be analyzed within the respiratory rate range, obtain the respiratory rate corresponding to the point to be analyzed, and divide the respiratory rate by a fixed value to obtain the actual frequency corresponding to the point to be analyzed. Divide the actual frequency by the fixed acquisition frequency and multiply by 2π to obtain the digital angular frequency θ of the point to be analyzed. Replace Z in the system function with... Obtain the frequency response of the corresponding point to be analyzed. ; Through formula The magnitude response of the point to be analyzed is calculated. ; Similarly, the amplitude response at different points to be analyzed within the respiratory rate range was calculated; The maximum value of the amplitude response is obtained by iterating through and comparing the amplitude responses of different analysis points, and the respiratory rate of the analysis point corresponding to the maximum value of the amplitude response is recorded as the estimated respiratory rate.

9. A method for generating respiratory waves and estimating respiratory rate in a wearable sensing system, characterized in that, A wearable sensing system applicable to any one of claims 1-8, the method comprising the following steps: Step S101: Process the electrocardiogram of the person to be tested to obtain the voltage intensity sequence, sequence length, and continuous acquisition duration of the electrocardiogram of the person to be tested. Step S102: Stack the voltage intensity sequence with convolution kernels in different convolutional layers to obtain the respiratory wave sequence of the person to be tested. Step S103: The respiratory wave sequence obtained by stacked convolution is judged to determine whether the obtained respiratory wave sequence is accurate; Step S104: Demodulate the respiratory wave of the person to be tested to obtain the final respiratory wave of the person to be tested; Step S105: Calculate the respiratory rate of the person being tested based on the final respiratory wave.

Images