Positioning detection method for respiration signals of large livestock based on millimeter wave radar

Abstract

The invention relates to the field of animal husbandry, in particular to a millimeter-wave radar-based positioning detection method for respiratory signals of large livestock. According to the invention, the adaptive angle estimation method based on the multi-channel real-value expansion matrix is adopted to realize angle estimation of the body area of the large livestock animal; secondly, according to extracted angle information and amplitude phase fusion, a most significant region of livestock respiratory movement is positioned; and finally, performing modal decomposition according to the signal energy corresponding to the selected angle interval, extracting a breathing mode in the positioning area, and reconstructing a breathing signal of the livestock. According to the FMCW radar-based large livestock respiratory signal detection method, a respiratory movement area is selected by fusing angle information with a weight factor, so that the universality and generalization of detection of respiratory signals of livestock of different varieties and different body sizes are improved, and the interference of livestock movement on the respiratory signals is reduced from the spatial dimension; and the extracted respiratory signal has relatively high accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent animal husbandry, in particular to a positioning detection method for respiratory signals of large livestock based on a millimeter wave radar. BACKGROUND

[0002] The health of livestock has always been a concern for people, and the physiological health of livestock is mainly checked regularly by artificial means. With the progress of science and technology, the physiological signals of livestock can be detected by sensors to learn about their health, and the health of livestock can be more closely monitored to improve the cure rate of livestock and the quality of livestock products and protect human health.

[0003] At present, the sensors for monitoring the physiological signals of livestock are divided into wearable and non-contact sensors. The wearable sensor records the respiratory frequency by measuring the temperature change near the nostrils of the livestock, and measures the respiratory rate by measuring the abdominal pressure change during exhalation and inhalation. The disadvantage is that wearing the device will cause discomfort, causing the sensor to slip off and affecting the measurement results. The limitations, inconvenience and inaccuracy of the wearable sensor have prompted the development of non-contact sensors.

[0004] The main way of non-contact monitoring of respiratory rate is to identify the mouth and nose as ROI based on a camera, detect the heat emitted by ROI when breathing, and determine the respiratory frequency by calculating the heat change when breathing. However, the camera is easily affected by the light environment, and the physiological signal features need to be extracted manually, which requires a large amount of data processing.

[0005] In contrast, the millimeter wave radar has the advantages of small size and being unaffected by environmental factors such as light, and can detect movements as small as a few tenths of a millimeter, which can be used to measure the slight vibrations caused by breathing. By capturing the reflected signal, the millimeter wave radar can determine the distance, speed and angle of the target, so as to detect the respiratory signal by calculating the vibration amplitude and frequency, and realize non-contact monitoring of vital signs.

[0006] There have been some achievements in the use of millimeter wave radar for non-contact measurement of animal vital signs at home and abroad, but there are still some deficiencies and problems to be solved, such as dynamic interference from natural noise, which causes distortion of the respiratory signal and affects the accuracy of respiratory monitoring; the non-stationary state of livestock affects the accuracy of radar detection; and the poor model generalization caused by the inconsistent respiratory characteristics of different livestock. SUMMARY

[0007] (I) Technical problems solved

[0008] In view of the deficiencies of the prior art, the present application provides a positioning detection method for respiratory signals of large livestock based on a millimeter wave radar, which solves the problems raised in the background art.

[0009] (II) Technical solutions

[0010] The present application specifically adopts the following technical solutions to achieve the above-mentioned purposes:

[0011] The positioning detection method for large livestock respiratory signals based on a millimeter wave radar comprises the following steps,

[0012] S1, an adaptive angle estimation method based on a multi-channel real value extension matrix is adopted, a signal is transmitted to a larger target through a millimeter wave radar array antenna, an intermediate frequency signal is obtained by mixing the received echo signal to obtain a bin file as original data, and then the I / Q components of the bin file data are extracted to form a multi-channel real value extension matrix as the input of angle prediction. The matrix is classified by feature extraction to realize DOA angle estimation;

[0013] S2, according to the DOA angle estimation obtained in S1, a method of positioning the micro-motion interval by fusing with the defined weight factor is used to complete the positioning of the most significant area of the livestock respiratory movement, and then the signal energy of the positioning area is subjected to variational mode decomposition to obtain the target respiratory signal.

[0014] Further, the S1 specific steps are as follows,

[0015] S11, the signal source under the large target belongs to a multi-signal source, the input multi-signal source IQ data matrix is spliced into a matrix by the real part and the imaginary part of the radar received signal, the IWR1843 radar is used for data acquisition, which contains 3 transmitting antennas and 4 receiving antennas, a total of 12 normal channels, and the matrix form is as follows:

[0016]

[0017] x ij (n) is the sampling value of the jth(j=1, 2, 3) transmitting signal of the ith(i=1, 2, 3, 4) receiving antenna, which is a complex number. Each complex number is decomposed into I / Q components and expanded into a real value matrix, which is a multi-channel real value extension matrix, and the matrix form is as follows:

[0018]

[0019] Wherein is the IQ data matrix of the multi-signal source, R(x ij (t)) and I(x ij (t)) are the real part and the imaginary part of the echo signal respectively, and the sampling time of each antenna is T(t=1, 2, …, T);

[0020] S12, the multi-channel real value extension matrix obtained by S11 is used as the input matrix M IQ, the network output is the DOA angle information, first generate the label vector, and discretize the field of view range [-60°, 60] into angle vectors of D (d = 1, 2,..., D) grid points

[0021]

[0022] θ d = -60° + (d-1)Δ θ

[0023] where Δ θ = 1° is the resolution, for multiple signal sources, the label vector y ∈ {0, 1} D , if there is a signal source at θ d , it is 1, otherwise it is 0, second, the network output is a D-dimensional vector O = [o1, o2,..., o D ], which is mapped to the probability by using the Sigmoid function:

[0024]

[0025] where p d represents the probability that the angle θ d exists a signal, and the output probability vector P = [p1,..., p D ];

[0026] S13, the loss function is the sum of binary cross-entropy of all angles:

[0027]

[0028] Minimize by backpropagation to adjust the network parameters W:

[0029]

[0030] S14, the network output probability vector P = [p1,..., p D ], select the angle whose p d ≥ τ by adaptive threshold τ:

[0031] τ = max(0.5, μ-1.5σ)

[0032] where μ and σ are the mean and standard deviation of the output probability, respectively, and the network output angle matrix

[0033]

[0034] Further, the S2 specific process is as follows:

[0035] S21. The angle matrix within the [-60°, 60°] field of view output by the S14 network. Divided into 4 non-overlapping angles, each angle ranging from 30°:

[0036] Γ k =[30k°,30(k+1)°),k∈{0,1,2,3}

[0037] Each angle interval Γ k Includes all angles that meet the above conditions;

[0038] S22, For each angle θ d Define the weighting factor w(θ) d ,t):

[0039]

[0040] Where, |x ij (θ d ,t)| represents the signal amplitude, φ(θ) d (t) represents the signal phase, and T represents the sampling time;

[0041] S23, for each interval Γ k Calculate the weighted aggregation signal

[0042]

[0043] For each Calculate its Fourier transform and the energy integral P. k :

[0044]

[0045] S24. Select the energy P that has the largest integral. * :

[0046] P * =max(P1,P2,P3,P4)

[0047] The maximum energy integral obtained is the region with the most significant respiratory movement in livestock. The weighted and aggregated signal within the corresponding interval is extracted. That is, x * (t), which includes respiratory signal components;

[0048] S25. The weighted and aggregated signal x within the selected interval... * (t) is used to perform empirical mode decomposition to obtain the frequency components of the respiratory signal.

[0049] (III) Beneficial Effects

[0050] Compared with the prior art, the present application provides a positioning detection method for large livestock respiratory signals based on millimeter wave radar, which has the following beneficial effects:

[0051] The present application solves the problems of inaccurate positioning, body movement and noise interference in large livestock respiratory signal detection, and can accurately position the abdominal position of large livestock due to body size problems, filter out interference caused by animal body movement from a spatial perspective, and solve the noise problem caused by small micro-movement, thereby improving the accuracy of respiratory signal detection. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The present application is a large livestock respiratory signal detection method based on FMCW radar, and the flowchart is shown in the figure.

[0053] Figure 2 The working principle and architecture of the present application are shown in the figure.

[0054] Figure 3 The livestock abdominal respiratory area selection flowchart of the present application is shown in the figure.

[0055] Figure 4 The respiratory signal extraction flowchart based on VMD algorithm of the present application is shown in the figure. DETAILED DESCRIPTION

[0056] The technical solutions of the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0057] EMBODIMENT

[0058] As shown in Figure 1 and Figure 2 , the present application is a positioning detection method for large livestock respiratory signals based on millimeter wave radar, which includes the following steps,

[0059] S1, as shown in Figure 3 , the radar array antenna receives the multi-channel real value expansion matrix composed of IQ components of the original signal as input, and extracts and classifies the matrix to estimate the DOA angle information. The present application uses IWR1843 radar for data collection, which contains 3 transmitting antennas and 4 receiving antennas, a total of 12 positive channels, and its matrix form is as follows:

[0060]

[0061] x ij (n) is the sample value of the jth (j = 1, 2, 3) transmitted signal of the ith (i = 1, 2, 3, 4) receiving antenna, complex form, each complex number is decomposed into I / Q components, extended to a real value matrix, which is a multi-channel real value extension matrix, the matrix form is as follows:

[0062]

[0063] where is the IQ data matrix of multiple signal sources, R(x ij (t)) and I(x ij (t)) are the real and imaginary parts of the echo signal respectively, and the sampling time of each antenna is T (t = 1, 2, …, T).

[0064] The obtained multi-channel real value extension matrix is taken as the input matrix M IQ of the prediction network, and the network output is the DOA angle information. First, a label vector is generated, and the field of view range [-60°, 60] is discretized into angle vectors of D (d = 1, 2, …, D) grid points

[0065]

[0066] θ d = -60° + (d-1)Δ θ

[0067] where Δ θ = 1° is the resolution. For multiple signal sources, the label vector y ∈ {0, 1} D if there is a signal source at θ d , otherwise 0. Secondly, the network output is a D-dimensional vector O = [o1, o2, …, o D ], which is mapped to the probability by using the Sigmoid function:

[0068]

[0069] where p d represents the probability of the angle θ d existing signal, and the output probability vector P = [p1, …, p D ].

[0070] The loss function is the sum of the binary cross-entropy of all angles:

[0071]

[0072] The network parameters W are adjusted by back propagation to minimize ​

[0073]

[0074] The probability vector P = [p1, …, pN] output by the network, select the angles with p D ≥ τ: d

[0075] τ = max(0.5, μ - 1.5σ)

[0076] where μ and σ are the mean and standard deviation of the output probabilities. The network output angles:

[0077]

[0078] S2, divide the angle matrix within the 120° field of view output by S1 into 4 non-overlapping angles, each with a range of 30°:

[0079] Γ k = [30k°, 30(k+1)°), k ∈ {0, 1, 2, 3}

[0080] Each interval Γ k contains all angles that satisfy the above conditions. For each angle θ d , define the weight factor w(θ d , t):

[0081]

[0082] where |x ij (θ d , t)| is the signal amplitude, φ(θ d , t) is the signal phase, and T is the sampling time. Then, for each interval Γ i , calculate the weighted aggregated signal and calculate the energy integral P k :

[0083]

[0084] Select the energy with the largest integral:

[0085] P * = max(P1, P2, P3, P4)

[0086] The largest energy integral obtained is the most significant area of livestock respiratory movement, and the weighted aggregated signal x * (t) in the corresponding interval is extracted.

[0087] According to the weighted aggregated signal x * ​​(t) as input of modal decomposition, as shown in Figure 4 First, initialize parameters, set the modal number K = 6, set the convergence factor α = 2000. Calculate the modal u k For each modal k, update the center frequency ω

[0088]

[0089] Update the center frequency ω k , the formula is as follows:

[0090]

[0091] Update the Lagrange multiplier λ, the formula is as follows:

[0092]

[0093] Where τ is the update step, And The Fourier transform of x * (t), u k (t) and λ(t). Select the center frequency ω k in the range of 0.2-1Hz, which is the breathing frequency of the animal, extract the corresponding modal u k (t) as the breathing signal.

[0094] Finally, it should be noted that: the above only for the preferred embodiments of the present application, and not for limiting the present application, although the foregoing embodiments of the present application are described in detail, for those skilled in the art, it still can be modified, or part of the technical features of the equivalent replacement. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the protection scope of the present application.

Claims

1. A method for locating and detecting respiratory signals in large livestock based on millimeter-wave radar, characterized in that: Includes the following steps, S1. An adaptive angle estimation method based on a multi-channel real-valued extended matrix is ​​adopted. The signal is transmitted to a large target through a millimeter-wave radar array antenna. The received echo signal is mixed to obtain an intermediate frequency signal as the original data bin file. Then, the I / Q components of the bin file data are extracted to form a multi-channel real-valued extended matrix as the input for angle prediction. The matrix is ​​then subjected to feature extraction and classification to achieve DOA angle estimation. S2. Based on the DOA angle estimate obtained in S1, the micro-motion interval is located by fusing it with the defined weighting factor. This completes the localization of the most significant respiratory movement region in livestock. Then, variational mode decomposition is performed on the signal energy of the localized region to obtain the respiratory signal of the target.

2. The method for locating and detecting respiratory signals of large livestock based on millimeter-wave radar according to claim 1, characterized in that: The specific steps of S1 are as follows: S11. The signal source under a large target is a multi-signal source. The input multi-signal source IQ data matrix is ​​formed by splicing the real and imaginary parts of the radar received signal into a matrix. Data acquisition is performed using the IWR1843 radar, which contains 3 transmitting antennas and 4 receiving antennas, for a total of 12 positive channels. Its matrix form is as follows: x ij (n) represents the sampled value of the j-th (j=1,2,3,4) transmitted signal from the i-th (i=1,2,3,4) receiving antenna. It is in complex form. Each complex number is decomposed into I / Q components and expanded into a real-valued matrix. It is a multi-channel real-valued extended matrix, and its matrix form is shown below: in For a multi-signal source IQ data matrix, R(x) ij (t)) and I(x ij (t) represents the real and imaginary parts of the echo signal, respectively, and the sampling time for each antenna is T (t = 1, 2, ..., T); S12. The multi-channel real-valued extended matrix obtained through S11 is used as the input matrix M of the prediction network. IQ The network output is DOA angle information. First, a label vector is generated, and the field of view [-60°, 60°] is discretized into angle vectors of D (d = 1, 2, ..., D) grid points. i d =-60°+(d-1)Δ θ Where, Δ θ =1° is the resolution. For multiple signal sources, the label vector y∈{0,1} D If at θ d The value is 1 if a signal source exists at a given location, and 0 otherwise. The network outputs a D-dimensional vector O = [ο1,ο2,...,ο]. D Using the Sigmoid function, the probability is mapped as follows: Where p d Represents angle θ d The probability of the presence of a signal, and the output probability vector P = [p1,…,p D ]; S13. The loss function is the sum of the binary cross-entropy from all angles: Minimize via backpropagation Adjust network parameter W: S14. The probability vector P output by the network is P = [p1,…,p D ], by using an adaptive threshold τ, select p d Angles ≥τ: τ = max(0.5, μ - 1.5σ) Where μ and σ are the mean and standard deviation of the output probabilities, respectively, and the network output angle matrix is ​​the network output angle matrix.

3. The method for locating and detecting respiratory signals of large livestock based on millimeter-wave radar according to claim 2, characterized in that: The specific process of S2 is as follows; S21. The angle matrix within the [-60°, 60°] field of view output by the S14 network. Divided into 4 non-overlapping angles, each angle ranging from 30°: Γ k =[30k°,30(k+1)°),k∈{0,1,2,3} Each angle interval Γ k Includes all angles that meet the above conditions; S22, For each angle θ d Define the weighting factor w(θ) d ,t): Where, |x ij (θ d ,t)| represents the signal amplitude, φ(θ) d (t) represents the signal phase, and T represents the sampling time; S23, for each interval Γ k Calculate the weighted aggregation signal For each Calculate its Fourier transform and the energy integral P. k : S24. Select the energy P that has the largest integral. * : P * =max(P1,P2,P3,P4) The maximum energy integral obtained is the region with the most significant respiratory movement in livestock. The weighted and aggregated signal within the corresponding interval is extracted. That is, x * (t), which includes respiratory signal components; S25. The weighted and aggregated signal x within the selected interval... * (t) is used to perform empirical mode decomposition to obtain the frequency components of the respiratory signal.

Images