Radar breathing signal detection method and system based on walking state

Abstract

The present invention discloses a radar breathing signal detection method and system based on walking state, and the method comprises the following steps: using Doppler radar to collect breathing signals, performing complex signal demodulation on the breathing signals, and obtaining a mixed signal of body motion signals and breathing signals; using an adaptive frequency shift detection algorithm to perform Doppler frequency shift elimination processing on the mixed signal of body motion signals and breathing signals, and obtaining a mixed signal of small body motion signals and breathing signals; using a deep neural network to perform small body motion removal processing on the mixed signal of small body motion signals and breathing signals, and obtaining a breathing signal; using a convolutional neural network to process the breathing signals, and obtaining a breathing frequency. The present invention can detect breathing signals during human walking, and compared with the existing radar breathing detection method, the application scenario is broader and the limitations are greatly reduced.

Description

Technical Field

[0001] The present invention belongs to the field of breathing detection, and in particular to a radar breathing signal detection method and system based on walking state. Background Art

[0002] Breathing is one of the important signals that can reflect a person's physical condition.

[0003] Existing radar breathing signal detection methods require the human body to be in a sitting state or to have only some minor body movements during the sitting process, such as swaying back and forth. Once the human body has large body movements, the measurement results will have a large deviation. However, no one has yet proposed a method for measuring breathing signals using a single radar while walking.

[0004] When walking, the signal detected by the radar is a mixed signal of body motion signal and breathing signal, among which there are large body motion signals such as Doppler frequency shift caused by human walking. In order to obtain the breathing signal from the mixed signal, it is necessary to first eliminate the influence of the large body motion signal to obtain a mixed signal of small body motion signal and breathing signal, and then remove the influence of small body motion to obtain the breathing signal.

[0005] Currently, no one has proposed a corresponding method to eliminate the gross body motion signal.

[0006] The existing algorithms for removing small body motion mainly include adaptive filtering and matched filtering. However, these methods require prior information when used. The adaptive filtering method requires an expected signal as input, and the matched filtering method requires a template signal as input. The application of the algorithms has great limitations. Summary of the invention

[0007] The purpose of the present invention is to address the shortcomings of existing radar breathing detection and provide a radar breathing signal detection method and system based on walking state, which only needs to use a single radar to detect the number of breaths per minute of a human body in the walking state.

[0008] The technical solution to achieve the purpose of the present invention is: a radar breathing signal detection method based on walking state, the method comprising the following steps:

[0009] Step 1, using Doppler radar to collect respiratory signals, and performing complex signal demodulation on the signals to obtain a mixed signal of body motion signals and respiratory signals;

[0010] Step 2, using an adaptive frequency shift detection algorithm to perform Doppler frequency shift elimination processing on the mixed signal of the body motion signal and the respiratory signal to obtain a mixed signal of the small body motion signal and the respiratory signal;

[0011] Step 3: Use a deep neural network to perform body movement removal processing on the mixed signal of body movement signals and respiratory signals to obtain respiratory signals;

[0012] Step 4: Use a convolutional neural network to process the respiratory signals to obtain the respiratory rate.

[0013] Further, in Step 2, the adaptive frequency shift detection algorithm is used to perform Doppler frequency shift cancellation processing on the mixed signal of body movement signals and respiratory signals to obtain the mixed signal of body movement signals and respiratory signals. Specifically:

[0014] Step 2-1: Set the window length t1 of the time window and the initial sliding window step size step = t1, in seconds. Perform sliding window processing on the mixed signal of body movement signals and respiratory signals. The time window is at the starting position of 0 s of the mixed signal of body movement signals and respiratory signals. Denote the sum of the frequencies corresponding to the maximum peak point and the second maximum peak point in the previous time window as Δ n-1 , whose initial value is 0. Denote the signal that needs to be processed for Doppler frequency shift cancellation obtained during the sliding window process as S, whose initial value is 0, and the output signal as Y, whose initial value is 0;

[0015] Step 2-2: Intercept the signal corresponding to the current time window from the mixed signal of body movement signals and respiratory signals, denote it as F, perform fast Fourier transform processing on it, find the frequencies corresponding to the maximum peak point and the second maximum peak point in the obtained frequency domain signal, and calculate their sum Δ n ;

[0016] Step 2-3: Judge the magnitudes of Δ n-1 and Δ n :

[0017] When |Δ n | > 0.1 and |Δ n-1 | ≤ 0.1, change the sliding window step size step to t2, where t2 < t1, and let S = [S, F];

[0018] When |Δ n | ≤ 0.1 and |Δ n-1 | > 0.1, perform Doppler frequency shift cancellation on S to obtain signal T. Denote the signal from the (t1 - t2) s to t1 s within the current time window as X, change the sliding window step size step to t1, clear signal S, and let Y = [Y, T, X];

[0019] When |Δ n-1 | > 0.1 and |Δ n | > 0.1, judge the magnitudes of Δ n-1 and Δ n . When Δ n-1 and Δ n>0, the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S = [S, X]. When Δ n-1 Δ n When <0, the Doppler frequency shift of S is eliminated to obtain signal T, and the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S=X, Y=[Y,T];

[0020] When |Δ n-1 |≤0.1 and |Δ n |≤0.1, let Y=[Y,F];

[0021] Step 2-4: the time window slides backward once with the sliding window step length step;

[0022] Step 2-5, repeating step 2-2, step 2-3 and step 2-4 until the time window slides to the end of the mixed signal of the body motion signal and the breathing signal;

[0023] Step 2-6, determine whether there is a signal in S. If so, perform Doppler frequency shift elimination on S to obtain signal T, set Y = [Y, T], and output Y; if not, directly output Y, and the output Y is a mixed signal of the small body movement signal and the breathing signal.

[0024] Furthermore, the Doppler frequency shift of S is eliminated in step 2-3 and step 2-6 to obtain a signal T, which is specifically:

[0025] Step 2-3-1, perform fast Fourier transform on signal S, find the point with the largest amplitude value from the corresponding frequency domain signal, and its corresponding frequency is the Doppler frequency shift value, recorded as f d ;

[0026] Step 2-3-2, generate a Doppler frequency shift compensation value with the same signal length as signal S, and its expression is:

[0027] A(n)=exp{-2jπf d n}

[0028] Where n = 1, 2, ..., N, n is the sampling point of signal S, N is the number of sampling points, and A(n) is the Doppler frequency shift compensation amount;

[0029] Step 2-3-3, multiply the signal S by the Doppler frequency shift compensation amount A(n) to obtain the signal T.

[0030] Furthermore, step 3 uses a deep neural network to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal, specifically:

[0031] Step 3-1, establish a data set, specifically: use the mixed signal of simulated breathing and small body movement as input, and use the corresponding simulated breathing signal as a label to create a training set and a validation set; the simulation model used for the simulated breathing signal is:

[0032]

[0033] Where r(t) represents the simulated breathing signal, a(t) represents the amplitude of the breathing signal, represents the phase of the respiratory signal, η(t) represents the noise, the parameter p0 is used to adjust the degree of nonlinear transformation, sgn represents the sign function, and the range of respiratory frequency is 0.1-0.5Hz;

[0034] Step 3-2, building a deep neural network model;

[0035] Step 3-3, training the deep neural network model, specifically: using the data set of the mixed signal of the small body movement signal and the respiratory signal established in step 3-1 to train the deep neural network model, to obtain a trained deep neural network model;

[0036] Step 3-4, restoring the respiratory signal, specifically: using the trained deep neural network model to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal.

[0037] Furthermore, step 4 uses a convolutional neural network to process the respiratory signal to obtain the respiratory frequency, specifically:

[0038] Step 4-1, establish a data set, specifically: use radar and breathing belt to simultaneously measure the breathing of m people in a sitting state, use the breathing signal measured by the radar as input, and extract the breathing frequency from the breathing signal measured by the breathing belt as a label to create a training set and a validation set;

[0039] Step 4-2, constructing a convolutional neural network model, specifically: establishing a convolutional neural network including M layers of one-dimensional convolutional layers;

[0040] Step 4-3, training the convolutional neural network model, specifically: using the respiratory signal data set established in step 4-1 to train the convolutional neural network model to obtain a trained convolutional neural network model;

[0041] Step 4-4, obtaining the respiratory frequency, specifically: using the trained convolutional neural network model to process the respiratory signal to obtain the respiratory frequency.

[0042] A radar breathing signal detection system based on walking state, the system comprising:

[0043] The signal acquisition module is used to collect the respiratory signal by using the Doppler radar, and perform complex signal demodulation on the respiratory signal to obtain a mixed signal of the body motion signal and the respiratory signal;

[0044] A first signal processing module is used to perform Doppler frequency shift elimination processing on a mixed signal of a body motion signal and a respiratory signal by using an adaptive frequency shift detection algorithm to obtain a mixed signal of a body motion signal and a respiratory signal;

[0045] The second signal processing module is used to use a deep neural network to remove the small body movement of the mixed signal of the small body movement signal and the respiratory signal to obtain the respiratory signal;

[0046] The respiratory rate detection module is used to process the respiratory signal using a convolutional neural network to obtain the respiratory rate.

[0047] Compared with the prior art, the present invention has the following significant advantages: 1) The present invention can detect respiratory signals during human walking. Compared with the existing radar respiratory detection method, the application scenario is broader and the limitations are greatly reduced; 2) A single radar can be used to detect respiratory signals in a walking state, and the equipment is simple and convenient for practical application; 3) When establishing a data set for a deep neural network, a new respiratory simulation model is used. Compared with the commonly used single harmonic model and periodic pulse model, this model is closer to the respiratory signal detected by the actual radar; 4) After the small body motion removal algorithm used in the present invention is trained, the interference of small body motion on the respiratory signal can be removed without any prior information.

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The figure is a flowchart of the radar breathing signal detection method based on walking state of the present invention.

[0050] Figure 2 FIG. 4 is a frequency domain diagram of a mixed signal of a body motion signal and a breathing signal detected in a walking state in one embodiment.

[0051] Figure 3 This is a comparison diagram of a breathing signal restored by an algorithm and a standard signal in a walking state in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] In addition, if the description in the present invention involves "first", "second", etc., the description of "first", "second", etc. is only for descriptive purposes and cannot be construed as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions between various embodiments may be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0054] Combined Figure 1 , the method for detecting radar respiration signals based on the walking state proposed by the present invention includes the following steps:

[0055] Step 1, collect respiration signals using a Doppler radar, perform complex signal demodulation on it, and obtain a mixed signal of body movement signals and respiration signals;

[0056] Step 2, use the adaptive frequency shift detection algorithm to perform Doppler frequency shift cancellation processing on the mixed signal of body movement signals and respiration signals to obtain a mixed signal of small body movement signals and respiration signals; specifically:

[0057] Step 2-1, set the window length t1 of the time window (preferably, t1 = 2), and the initial sliding window step size step = t1, with the unit of s, perform sliding window processing on the mixed signal of body movement signals and respiration signals, the time window is at the starting 0 s position of the mixed signal of body movement signals and respiration signals, record the sum of the frequencies corresponding to the maximum peak point and the second maximum peak point in the previous time window as Δ n-1 , its initial value is 0, the signal to be subjected to Doppler frequency shift cancellation processing obtained during the sliding window process is denoted as S, its initial value is 0, and the output signal is denoted as Y, its initial value is 0;

[0058] Step 2-2, intercept the signal corresponding to the current time window from the mixed signal of body movement signals and respiration signals, denote it as F, perform fast Fourier transform processing on it, find the frequencies corresponding to the maximum peak point and the second maximum peak point in the obtained frequency domain signal, and calculate the sum of the two as Δ n ;

[0059] Step 2-3, judge the magnitudes of Δ n-1 and Δ n :

[0060] When |Δ n | > 0.1 and |Δ n-1 | ≤ 0.1, change the sliding window step size step to t2 (preferably, t2 = 0.5), t2 < t1, and let S = [S, F];

[0061] When |Δ n |≤0.1 and |Δ n-1 When |>0.1, the Doppler frequency shift of S is eliminated to obtain signal T. The signal from (t1-t2)s to t1s (1.5 to 2s) in the current time window is recorded as X. The sliding window step is changed to t1, the signal S is cleared, and Y=[Y,T,X];

[0062] When |Δ n-1 |>0.1 and |Δ n |>0.1, judge Δ n-1 Δ n The size of Δ n-1 Δ n >0, the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S = [S, X]. When Δ n-1 Δ n When <0, the Doppler frequency shift of S is eliminated to obtain signal T, and the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S=X, Y=[Y,T];

[0063] When |Δ n-1 |≤0.1 and |Δ n |≤0.1, let Y=[Y,F];

[0064] Here, the Doppler frequency shift of S is eliminated to obtain the signal T, which is specifically:

[0065] Step 2-3-1, perform fast Fourier transform on signal S, find the point with the largest amplitude value from the corresponding frequency domain signal, and its corresponding frequency is the Doppler frequency shift value, recorded as f d ;

[0066] Step 2-3-2, generate a Doppler frequency shift compensation value with the same signal length as signal S, and its expression is:

[0067] A(n)=exp{-2jπf d n}

[0068] Where n = 1, 2, ..., N, n is the sampling point of signal S, N is the number of sampling points, and A(n) is the Doppler frequency shift compensation amount;

[0069] Step 2-3-3, multiplying the signal S by the Doppler frequency shift compensation amount A(n) to obtain the signal T;

[0070] Step 2-4: the time window slides backward once with the sliding window step length step;

[0071] Step 2-5, repeating step 2-2, step 2-3 and step 2-4 until the time window slides to the end of the mixed signal of the body motion signal and the breathing signal;

[0072] Step 2-6, determine whether there is a signal in S. If so, perform Doppler frequency shift elimination on S to obtain signal T, set Y = [Y, T], and output Y; if not, directly output Y, and the output Y is a mixed signal of the small body movement signal and the breathing signal.

[0073] Step 3: Use a deep neural network to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal; specifically:

[0074] Step 3-1, establish a data set, specifically: use the mixed signal of simulated breathing and small body movement as input, and use the corresponding simulated breathing signal as a label to create a training set and a validation set; the simulation model used for the simulated breathing signal is:

[0075]

[0076] Where r(t) represents the simulated breathing signal, a(t) represents the amplitude of the breathing signal, represents the phase of the respiratory signal, η(t) represents the noise, the parameter p0 is used to adjust the degree of nonlinear transformation, sgn represents the sign function, and the range of respiratory frequency is 0.1-0.5Hz;

[0077] Step 3-2, constructing a deep neural network model; according to the length and characteristics of the mixed signal of the small body movement signal and the respiratory signal of the present invention, the optimal model parameters are selected as follows: the input layer size is 1000, the number of hidden layers is 5, the sizes of each layer are 400, 160, 64, 160, 400 respectively, and the size of the output layer is 1000;

[0078] Step 3-3, training the deep neural network model, specifically: using the data set of the mixed signal of the small body movement signal and the respiratory signal established in step 3-1 to train the deep neural network model, to obtain a trained deep neural network model;

[0079] Step 3-4, restoring the respiratory signal, specifically: using the trained deep neural network model to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal.

[0080] Step 4: Use convolutional neural network to process the respiratory signal and obtain the respiratory frequency. Specifically:

[0081] Step 4-1, establish a data set, specifically: use radar and breathing belt to simultaneously measure the breathing of m people in a sitting state, use the breathing signal measured by the radar as input, and use the breathing frequency extracted from the breathing signal measured by the breathing belt as a label to create a training set and a validation set;

[0082] Step 4-2, constructing a convolutional neural network model, specifically: establishing a convolutional neural network including M layers of one-dimensional convolutional layers; according to the length and characteristics of the respiratory signal of the present invention, the optimal model parameters are selected as follows: the number of one-dimensional convolutional layers is 3, the output channels of each layer are 64, the convolution kernel size is 5, the convolution step size is 1, and the edge padding is 2;

[0083] Step 4-3, training the convolutional neural network model, specifically: using the respiratory signal data set established in step 4-1 to train the convolutional neural network model to obtain a trained convolutional neural network model;

[0084] Step 4-4, obtaining the respiratory frequency, specifically: using the trained convolutional neural network model to process the respiratory signal to obtain the respiratory frequency.

[0085] The present invention provides a radar breathing signal detection system based on walking state, the system comprising the following steps executed in sequence:

[0086] The signal acquisition module is used to collect the respiratory signal by using the Doppler radar, and perform complex signal demodulation on the respiratory signal to obtain a mixed signal of the body motion signal and the respiratory signal;

[0087] The first signal processing module is used to perform Doppler frequency shift elimination processing on the mixed signal of the body motion signal and the respiratory signal by using an adaptive frequency shift detection algorithm to obtain a mixed signal of the body motion signal and the respiratory signal; the module includes the following steps executed in sequence:

[0088] The sliding window setting and processing unit is used to set the window length t1 of the time window and the initial sliding window step length step=t1, the unit is s, and the sliding window processing is performed on the mixed signal of the body motion signal and the respiratory signal. The time window is at the starting position 0s of the mixed signal of the body motion signal and the respiratory signal, and the sum of the corresponding frequencies of the maximum peak point and the second largest peak point in the previous time window is recorded as Δ n-1 , its initial value is 0, the signal that needs to be processed for Doppler frequency shift elimination determined during the sliding window process is recorded as S, its initial value is 0, and the output signal is recorded as Y, its initial value is 0;

[0089] The first calculation unit is used to intercept the signal at the corresponding position of the current time window from the mixed signal of the body motion signal and the respiratory signal, denoted as F, perform fast Fourier transform processing on it, find the frequency corresponding to the maximum peak point and the second largest peak point on the obtained frequency domain signal, and calculate the sum of the two Δ n ;

[0090] A judgment unit, used to judge Δ n-1 and Δ n in terms of magnitude:

[0091] When |Δ n | > 0.1 and |Δ n-1 | ≤ 0.1, change the sliding window step size step to t2, where t2 < t1, and let S = [S, F];

[0092] When |Δ n | ≤ 0.1 and |Δ n-1 | > 0.1, perform Doppler frequency shift cancellation on S to obtain signal T, record the signal from (t1 - t2)s to t1 s within the current time window as X, change the sliding window step size step to t1, clear signal S, and let Y = [Y, T, X];

[0093] When |Δ n-1 | > 0.1 and |Δ n | > 0.1, judge the magnitude of Δ n-1 Δ n When Δ n-1 Δ n > 0, record the signal from (t1 - t2)s to t1 s within the current time window as X, and let S = [S, X]. When Δ n-1 Δ n < 0, perform Doppler frequency shift cancellation on S to obtain signal T, record the signal from (t1 - t2)s to t1 s within the current time window as X, let S = X, and Y = [Y, T];

[0094] When |Δ n-1 | ≤ 0.1 and |Δ n | ≤ 0.1, let Y = [Y, F];

[0095] A window sliding unit, used to slide the time window backward once with the sliding window step size step;

[0096] A loop termination unit, used to repeat the first calculation unit to the window sliding unit until the time window slides to the end of the mixed signal of the body movement signal and the breathing signal;

[0097] A mixed signal acquisition unit of the small body movement signal and the breathing signal, used to judge whether there is a signal in S. If there is, perform Doppler frequency shift cancellation on S to obtain signal T, let Y = [Y, T], and output Y; if not, directly output Y, and the output Y is the mixed signal of the small body movement signal and the breathing signal.

[0098] A second signal processing module, used to perform small body movement removal processing on the mixed signal of the small body movement signal and the breathing signal by using a deep neural network to obtain a breathing signal; this module includes the following steps executed in sequence:

[0099] The first construction unit is used to establish a data set, specifically: using the mixed signal of simulated breathing and small body movement as input, and corresponding to the simulated breathing signal as a label to create a training set and a verification set; wherein the simulation model used by the simulated breathing signal is:

[0100]

[0101] Where r(t) represents the simulated breathing signal, a(t) represents the amplitude of the breathing signal, represents the phase of the respiratory signal, η(t) represents the noise, the parameter p0 is used to adjust the degree of nonlinear transformation, sgn represents the sign function, and the range of respiratory frequency is 0.1-0.5Hz;

[0102] The second construction unit is used to construct a deep neural network model;

[0103] The first model training unit is used to train the deep neural network model, specifically: using the established data set of mixed signals of small body movement signals and respiratory signals to train the deep neural network model to obtain a trained deep neural network model;

[0104] The respiratory signal recovery unit is used to recover the respiratory signal, specifically: using the trained deep neural network model to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal.

[0105] The respiratory rate detection module is used to process the respiratory signal using a convolutional neural network to obtain the respiratory rate. This module includes the following steps:

[0106] The third construction unit is used to establish a data set, specifically: using a radar and a breathing belt to simultaneously measure the breathing of m people in a sitting state, taking the breathing signal measured by the radar as input, and extracting the breathing frequency from the breathing signal measured by the breathing belt as a label to create a training set and a validation set;

[0107] The fourth construction unit is used to construct a convolutional neural network model, specifically: establish a convolutional neural network including M layers of one-dimensional convolutional layers;

[0108] The second model training unit trains the convolutional neural network model, specifically: using the respiratory signal data set established in step 4-1 to train the convolutional neural network model to obtain a trained convolutional neural network model;

[0109] The respiratory frequency extraction unit is used to obtain the respiratory frequency, specifically: using the trained convolutional neural network model to process the respiratory signal to obtain the respiratory frequency.

[0110] For the specific definition of the radar breathing signal detection system based on walking state, please refer to the definition of the radar breathing signal detection method based on walking state in the above text, which will not be repeated here. Each module in the above-mentioned radar breathing signal detection system based on walking state can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0111] The present invention is further described in detail below in conjunction with embodiments.

[0112] Example

[0113] According to step 1, a mixed signal of a 10s period of gross body movement signal and respiratory signal is obtained, with a sampling frequency of 100Hz. At the same time, the respiratory signal of the subject is detected by a respiratory belt as a reference signal. The corresponding frequency domain is as follows Figure 2 As shown by Figure 2 It can be seen that due to the presence of large body movement and small body movement, the respiratory frequency cannot be directly obtained. The Doppler frequency shift of the mixed signal of large body movement signal and respiratory signal is eliminated by using the adaptive frequency shift detection algorithm, and the mixed signal of small body movement signal and respiratory signal is processed by using the deep neural network to obtain the respiratory signal, such as Figure 3 As shown by Figure 3 It can be seen that the obtained breathing signal is consistent with the reference signal. The parameters of the deep neural network are as follows:

[0114] Input layer size: 1000;

[0115] First hidden layer size: 400;

[0116] Second hidden layer size: 160;

[0117] Third hidden layer size: 64;

[0118] Fourth hidden layer size: 160;

[0119] Fifth hidden layer size: 400;

[0120] Output layer size: 1000;

[0121] Learning rate: 0.0001;

[0122] Batch size: 20

[0123] Number of training rounds: 50.

[0124] The convolutional neural network is used to process the respiratory signal to obtain the respiratory frequency. The parameters of the convolutional neural network are as follows:

[0125] Input layer parameters: input channel size is 1, number of convolution channels is 64, convolution kernel size is 5, convolution step size is 1, and edge padding is 2;

[0126] The first hidden layer parameters are: input channel size is 64, number of convolution channels is 64, convolution kernel size is 5, convolution step size is 1, and edge padding is 2;

[0127] Second hidden layer parameters: input channel size is 64, number of convolution channels is 64, convolution kernel size is 5, convolution step size is 1, edge padding is 2;

[0128] Output layer parameters: input size is 64*1000, output size is 1000;

[0129] Learning rate: 0.001;

[0130] Batch size: 32

[0131] Number of training rounds: 50.

[0132] 10 packets of actual human walking measurement data were collected, and the results after processing by this algorithm were compared with the results measured by the breathing belt, as shown in Table 1 below.

[0133] Table 1 Comparison of respiratory rate

[0134]

[0135]

[0136] As shown in Table 1, the average error rate of the 10 packets of data is 8.16%, which is relatively small. It can be verified that the present invention can detect the breathing frequency using a single radar during human movement, and is effective and feasible.

[0137] The present invention can detect breathing signals while a person is walking. Compared with the existing radar breathing detection method, the present invention has a wider application scenario and greatly reduces limitations.

[0138] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A radar breathing signal detection method based on walking state, characterized in that: The method comprises the following steps: Step 1, using Doppler radar to collect respiratory signals, and performing complex signal demodulation on the signals to obtain a mixed signal of body motion signals and respiratory signals; Step 2, using an adaptive frequency shift detection algorithm to perform Doppler frequency shift elimination processing on the mixed signal of the body motion signal and the respiratory signal to obtain a mixed signal of the small body motion signal and the respiratory signal; Step 3, using a deep neural network to remove small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain a respiratory signal; Step 4: Use a convolutional neural network to process the respiratory signal and obtain the respiratory frequency.

2. The radar breathing signal detection method based on walking state according to claim 1 is characterized in that: Step 2 uses an adaptive frequency shift detection algorithm to perform Doppler frequency shift elimination processing on the mixed signal of the body motion signal and the respiratory signal to obtain a mixed signal of the body motion signal and the respiratory signal, specifically: Step 2-1, set the window length t1 of the time window and the initial sliding window step length step = t1, the unit is s, and perform sliding window processing on the mixed signal of the body motion signal and the respiratory signal. The time window is at the starting position of the mixed signal of the body motion signal and the respiratory signal at 0s. The sum of the corresponding frequencies of the maximum peak point and the second largest peak point in the previous time window is recorded as Δ n-1 , its initial value is 0, the signal that needs to be processed for Doppler frequency shift elimination determined during the sliding window process is recorded as S, its initial value is 0, and the output signal is recorded as Y, its initial value is 0; Step 2-2, intercept the signal at the corresponding position of the current time window from the mixed signal of the body motion signal and the respiratory signal, record it as F, perform fast Fourier transform processing on it, find the frequency corresponding to the maximum peak point and the second largest peak point in the obtained frequency domain signal, and calculate the sum of the two Δ n ; Step 2-3, determine Δ n-1 With Δ n Size: When |Δ n | > 0.1 and |Δ n-1 | ≤ 0.1, change the sliding window step size step to t2, where t2 < t1, and let S = [S, F]; When |Δ n |≤0.1 and |Δ n-1 When |>0.1, the Doppler frequency shift of S is eliminated to obtain signal T. The signal from (t1-t2)s to t1s in the current time window is recorded as X. The sliding window step length step is changed to t1, the signal S is cleared, and Y=[Y,T,X]; When |Δ n-1 |>0.1 and |Δ n |>0.1, judge Δ n-1 Δ n The size of Δ n-1 Δ n >0, the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S = [S, X]. When Δ n-1 Δ n When <0, the Doppler frequency shift of S is eliminated to obtain signal T, and the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S=X, Y=[Y,T]; When |Δ n-1 |≤0.1 and |Δ n |≤0.1, let Y=[Y,F]; Step 2-4: the time window slides backward once with the sliding window step length step; Step 2-5, repeating step 2-2, step 2-3 and step 2-4 until the time window slides to the end of the mixed signal of the body motion signal and the breathing signal; Step 2-6, determine whether there is a signal in S. If so, perform Doppler frequency shift elimination on S to obtain signal T, set Y = [Y, T], and output Y; if not, directly output Y, and the output Y is a mixed signal of the small body movement signal and the breathing signal.

3. The radar breathing signal detection method based on walking state according to claim 2 is characterized in that: The Doppler frequency shift of S is eliminated as described in step 2-3 and step 2-6 to obtain a signal T, which is specifically: Step 2-3-1, perform fast Fourier transform on signal S, find the point with the largest amplitude value from the corresponding frequency domain signal, and its corresponding frequency is the Doppler frequency shift value, recorded as f d ; Step 2-3-2, generate a Doppler frequency shift compensation value with the same signal length as signal S, and its expression is: A(n)=exp{-2jπf d n} Where n = 1, 2, ..., N, n is the sampling point of signal S, N is the number of sampling points, and A(n) is the Doppler frequency shift compensation amount; Step 2-3-3, multiply the signal S by the Doppler frequency shift compensation amount A(n) to obtain the signal T.

4. The radar breathing signal detection method based on walking state according to claim 3 is characterized in that: Step 3 uses a deep neural network to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal, specifically: Step 3-1, establish a data set, specifically: use the mixed signal of simulated breathing and small body movement as input, and use the corresponding simulated breathing signal as a label to create a training set and a validation set; the simulation model used for the simulated breathing signal is: Where r(t) represents the simulated breathing signal, a(t) represents the amplitude of the breathing signal, represents the phase of the respiratory signal, η(t) represents the noise, the parameter p0 is used to adjust the degree of nonlinear transformation, sgn represents the sign function, and the range of respiratory frequency is 0.1-0.5Hz; Step 3-2, building a deep neural network model; Step 3-3, training the deep neural network model, specifically: using the data set of the mixed signal of the small body movement signal and the respiratory signal established in step 3-1 to train the deep neural network model, to obtain a trained deep neural network model; Step 3-4, restoring the respiratory signal, specifically: using the trained deep neural network model to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal.

5. The radar breathing signal detection method based on walking state according to claim 4 is characterized in that: The parameters of the deep neural network model described in step 3-2 include: the input layer size is 1000, the number of hidden layers is 5, the sizes of each layer are 400, 160, 64, 160, 400 respectively, and the size of the output layer is 1000.

6. The radar breathing signal detection method based on walking state according to claim 5 is characterized in that: Step 4 uses a convolutional neural network to process the respiratory signal to obtain the respiratory frequency, specifically: Step 4-1, establish a data set, specifically: use radar and breathing belt to simultaneously measure the breathing of m people in a sitting state, use the breathing signal measured by the radar as input, and use the breathing frequency extracted from the breathing signal measured by the breathing belt as a label to create a training set and a validation set; Step 4-2, constructing a convolutional neural network model, specifically: establishing a convolutional neural network including M layers of one-dimensional convolutional layers; Step 4-3, training the convolutional neural network model, specifically: using the respiratory signal data set established in step 4-1 to train the convolutional neural network model to obtain a trained convolutional neural network model; Step 4-4, obtaining the respiratory frequency, specifically: using the trained convolutional neural network model to process the respiratory signal to obtain the respiratory frequency.

7. The radar breathing signal detection method based on walking state according to claim 6 is characterized in that: The parameters of the convolutional neural network model described in step 4-2 include: the number of one-dimensional convolutional layers is 3, the output channels of each layer are 64, the convolution kernel size is 5, the convolution step size is 1, and the edge padding is 2.

8. A radar breathing signal detection system based on walking state that implements the method described in any one of claims 1 to 7, characterized in that: The system includes executing in sequence: The signal acquisition module is used to collect the respiratory signal by using the Doppler radar, and perform complex signal demodulation on the respiratory signal to obtain a mixed signal of the body motion signal and the respiratory signal; A first signal processing module is used to perform Doppler frequency shift elimination processing on a mixed signal of a body motion signal and a respiratory signal by using an adaptive frequency shift detection algorithm to obtain a mixed signal of a body motion signal and a respiratory signal; The second signal processing module is used to use a deep neural network to remove the small body movement of the mixed signal of the small body movement signal and the respiratory signal to obtain the respiratory signal; The respiratory rate detection module is used to process the respiratory signal using a convolutional neural network to obtain the respiratory rate.

9. The radar breathing signal detection system based on walking state according to claim 8, characterized in that: The first signal processing module includes the following steps executed in sequence: The sliding window setting and processing unit is used to set the window length t1 of the time window and the initial sliding window step length step=t1, the unit is s, and the sliding window processing is performed on the mixed signal of the body motion signal and the respiratory signal. The time window is at the starting position 0s of the mixed signal of the body motion signal and the respiratory signal, and the sum of the corresponding frequencies of the maximum peak point and the second largest peak point in the previous time window is recorded as Δ n-1 , its initial value is 0, the signal that needs to be processed for Doppler frequency shift elimination determined during the sliding window process is recorded as S, its initial value is 0, and the output signal is recorded as Y, its initial value is 0; The first calculation unit is used to intercept the signal at the corresponding position of the current time window from the mixed signal of the body motion signal and the respiratory signal, denoted as F, perform fast Fourier transform processing on it, find the frequency corresponding to the maximum peak point and the second largest peak point on the obtained frequency domain signal, and calculate the sum of the two Δ n ; A judging unit is used to judge Δ n-1 With Δ n Size: When |Δ n | > 0.1 and |Δ n-1 | ≤ 0.1, change the sliding window step size step to t2, where t2 < t1, and let S = [S, F]; When |Δ n |≤0.1 and |Δ n-1 When |>0.1, the Doppler frequency shift of S is eliminated to obtain signal T. The signal from (t1-t2)s to t1s in the current time window is recorded as X. The sliding window step length step is changed to t1, the signal S is cleared, and Y=[Y,T,X]; When |Δ n-1 |>0.1 and |Δ n |>0.1, judge Δ n-1 Δ n The size of Δ n-1 Δ n >0, the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S = [S, X]. When Δ n-1 Δ n When <0, the Doppler frequency shift of S is eliminated to obtain signal T, and the signal from (t1-t2)s to t1s in the current time window is recorded as X, and S=X, Y=[Y,T]; When |Δ n-1 |≤0.1 and |Δ n |≤0.1, let Y=[Y,F]; The window sliding unit is used to slide the time window backward once with the sliding window step size step; A loop termination unit, used for repeating the first calculation unit to the window sliding unit until the time window slides to the end of the mixed signal of the body motion signal and the breathing signal; The mixed signal acquisition unit of the small body movement signal and the breathing signal is used to determine whether there is a signal in S. If so, the Doppler frequency shift of S is eliminated to obtain the signal T, and Y=[Y,T] is set, and Y is output; if not, Y is directly output, and the output Y is the mixed signal of the small body movement signal and the breathing signal.

10. The radar breathing signal detection system based on walking state according to claim 9, characterized in that: The second signal processing module includes the following steps: The first construction unit is used to establish a data set, specifically: using the mixed signal of simulated breathing and small body movement as input, and corresponding to the simulated breathing signal as a label to create a training set and a verification set; wherein the simulation model used by the simulated breathing signal is: Where r(t) represents the simulated breathing signal, a(t) represents the amplitude of the breathing signal, represents the phase of the respiratory signal, η(t) represents the noise, the parameter p0 is used to adjust the degree of nonlinear transformation, sgn represents the sign function, and the range of respiratory frequency is 0.1-0.5Hz; The second construction unit is used to construct a deep neural network model; The first model training unit is used to train the deep neural network model, specifically: using the established data set of mixed signals of small body movement signals and respiratory signals to train the deep neural network model to obtain a trained deep neural network model; The respiratory signal recovery unit is used to recover the respiratory signal, specifically: using the trained deep neural network model to remove the small body motion from the mixed signal of the small body motion signal and the respiratory signal to obtain the respiratory signal.

Images