Heart rate detection method based on exercise power

Abstract

The invention relates to the technical field of heart rate detection, in particular to a heart rate detection method based on motion power, which comprises the following steps: acquiring radar signals and heart rate signals in different motion states; performing frequency spectrum extraction of different distance values on the radar signal, calculating the volume of the body part by using the area of the body part, and obtaining the mass of the body part through the volume of the body part; calculating the displacement of the body part; calculating the speed and acceleration of the body part; calculating the physical force of the body part by using the acceleration; calculating the physical workload of the body part according to the physical force; converting the physical workload into motion power; constructing a first LSTM model, and inputting the exercise power and the actually-acquired heart rate into the first LSTM model to obtain a first predicted heart rate; and constructing a second LSTM model by taking the exercise power and the first predicted heart rate as input, and outputting a second predicted heart rate. The heart rate estimation based on the density method is easily influenced by factors such as individual differences, motion types and intensity.

Description

Technical Field

[0001] The present invention relates to the technical field of heart rate detection, and in particular, to a heart rate detection method based on exercise power. Background Art

[0002] Heart rate detection plays multiple roles in health management and medical diagnosis. By regularly monitoring the heart rate, potential health problems can be detected in a timely manner and corresponding measures can be taken for intervention and treatment.

[0003] Existing heart rate detections include contact heart rate detection, but it requires human contact measurement through wearable devices.

[0004] The powerful potential of millimeter-wave radar-based heart rate detection technology as a non-contact monitoring tool; such as camera-guided frequency-modulated continuous wave (FMCW) radar and Doppler radar for monitoring heart rate; combined with RGB-D cameras and thermal cameras, which can achieve non-contact detection of multiple people's heart rates in public places. However, the accuracy of thermal cameras is greatly affected by the environment and there are privacy issues.

[0005] For the waveform corresponding to the heart rate signal in the patent with the publication number CN111481184A, a density-based method is used to estimate the heart rate, all peaks in the heart rate spectrum are obtained, and the accumulated peaks are divided into multiple clusters using the dBscan clustering algorithm; however, affected by factors such as individual differences and the type and intensity of exercise, the accuracy of the density method needs to be improved; in addition, the density method is too dependent on the analysis and processing of historical data and has poor real-time performance.

[0006] Therefore, how to improve the accuracy of heart rate detection in a moving state is a problem that needs to be continuously solved. Summary of the Invention

[0007] Aiming at the deficiencies of the existing methods, the present invention uses the received power of the body reflection part to evaluate the human body weight, combines the displacement of the body part to estimate the human exercise power under different actions, and uses the LSTM network to model the exercise power and heart rate to achieve accurate prediction of the heart rate.

[0008] The technical solution adopted by the present invention is: a heart rate detection method based on exercise power includes the following steps:

[0009] Step 1: Collect radar signals in different exercise states and corresponding heart rate signals.

[0010] As a preferred embodiment of the present invention, the exercise states include: rope skipping, stationary cycling, and running on a fitness machine.

[0011] As a preferred embodiment of the present invention, the radar signals include: FMCW radar.

[0012] As a preferred embodiment of the present invention, the heart rate signal is collected by a wearable device.

[0013] Step 2: Extract the spectra of the radar signals at different distance values, calculate the volume of the body part using the area of the body part, and obtain the mass of the body part from the volume of the body part;

[0014] As a preferred embodiment of the present invention, the formula for the mass of the body part is:

[0015] m k = ρμD k C k (4)

[0016] where ρ represents the mass density, μ is the proportionality coefficient, C k is the received power of the k-distance bin, and D k is the proportionality coefficient for converting the received power C k to the area of the body part in the k-distance bin.

[0017] Step 3: Calculate the displacement of the body part; calculate the acceleration of the body part; calculate the physical force using the acceleration and mass of the body part; calculate the physical work done by the body part based on the physical force; convert the physical work done to the exercise power;

[0018] As a preferred embodiment of the present invention, the formula for the physical work done is:

[0019]

[0020] where F k = m k a k , a k is the acceleration, and Δd is the displacement of the body part.

[0021] As a preferred embodiment of the present invention, the formula for the displacement of the body part is:

[0022]

[0023] where φ k is the phase difference between two adjacent frames, and λ is the wavelength.

[0024] As a preferred embodiment of the present invention, the formula for converting the physical work done to the exercise power is:

[0025]

[0026] where T f is the period of the radar frame.

[0027] Step 4: Construct the first LSTM model, and input the exercise power and the actually measured heart rate into the first LSTM model to obtain the first predicted heart rate;

[0028] As a preferred embodiment of the present invention, the first LSTM model is a 2-unit LSTM composed of dense layers.

[0029] Step 5: Use the exercise power and the first predicted heart rate as inputs to construct a second LSTM model, and output the second predicted heart rate;

[0030] As a preferred embodiment of the present invention, the second LSTM model includes: a first LSTM layer, a Dropout layer, a second LSTM layer, and a fully connected layer in cascade.

[0031] Advantages of the present invention:

[0032] 1. The present invention calculates the part quality through the received power of the body reflection part, calculates the physical workload using the displacement, acceleration, and physical force of the body part, converts the physical workload into exercise power, and constructs the relationship between exercise power and heart rate, which is simpler than the existing radar signal method;

[0033] 2. Construct two LSTM models to accurately predict the heart rate value through secondary prediction. Description of the Drawings

[0034] Figure 1 is a schematic diagram of the heart rate detection method based on exercise power of the present invention. Detailed Embodiments

[0035] The present invention will be further described below in conjunction with the drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, so it only shows the components related to the present invention.

[0036] As Figure 1 shown, a heart rate detection method based on exercise power includes the following steps:

[0037] Step 1: Collect radar signals in different exercise states and the corresponding heart rate signals;

[0038] The exercise states include: rope skipping, stationary bicycle, running on a fitness machine, etc.

[0039] The radar signals include: FMCW radar.

[0040] The heart rate signals can be obtained through wearable devices, such as sports bracelets, smart watches, electrocardiogram monitors, contact heart rate sensors: such as BXT16, BXT128, BXT188, etc.

[0041] Step 2: extract the spectrum of the radar signal at different distance values, calculate the volume of the body part using the area of ​​the body part, and obtain the mass of the body part through the volume of the body part;

[0042] Different distance values ​​are the distance FFTs corresponding to different reflection points in the target scene. Since the reflected waves from equidistant body parts have the same phase, the received power of the reflected waves is superimposed; therefore, the received power of each distance FFT bin is related to the body area S at different distance values. k Linearly proportional; first calculate the area S of the body part at different distance values k , the formula is:

[0043] S k =D k C k (1)

[0044] Among them, C k is the received power of the k-distance bin, D k To receive power C k Scaling factor for conversion to body part area in k distance bins.

[0045] Assume that the volume of the body part is V k Relative to S k Linear, calculate V k , the formula is:

[0046] V k =μS k (2)

[0047] Where μ is the proportionality coefficient.

[0048] Assuming that the thickness of different parts of the body is similar, the formula for obtaining the mass of the body part is:

[0049] m k =ρV k (3)

[0050] Here, ρ represents the mass density.

[0051] The formula for body part mass obtained by using formula 1-3 is:

[0052] m k =ρμD k C k (4)

[0053] In practice, the absolute value of the mass m of a body part does not matter. k , because it is proportional to the reflected power C k , using C k Predict body part mass for subsequent movement power calculation.

[0054] Step 3: Calculate the displacement of the body part; calculate the speed and acceleration of the body part; calculate the physical force of the body part using the acceleration; calculate the physical work of the body part based on the physical force; convert the physical work into motion power;

[0055] The displacement formula of the body part is:

[0056]

[0057] where φ k is the phase difference between two adjacent frames, and λ is the wavelength.

[0058] The formulas for speed and acceleration are:

[0059]

[0060]

[0061] where T f is the period of the radar frame.

[0062] Calculate the physical force of the body part. The formula is:

[0063] F k = m k a k (8)

[0064] Since the radar frame is usually very short, on the order of milliseconds, assume that the force driving the body movement remains constant within the period; therefore, the formula for the physical work of the body part within the k-max(k) range bin and the i-max(i) frame is:

[0065]

[0066] where i represents the frame.

[0067] Obtain the total physical work of the body part. The formula is:

[0068]

[0069] where M is the total number of range bins.

[0070] Convert the physical work into motion power. The formula is:

[0071]

[0072] Construct a dataset of power data and heart rate data for subsequent model prediction.

[0073] Step 4: Construct the first LSTM model, input the exercise power and the actual measured heart rate into the first LSTM model to obtain the first predicted heart rate;

[0074] The first LSTM model consists of a 2-unit LSTM with a dense layer having a kernel size of 128, and uses Stochastic Gradient Descent (SGD) with a weight decay of 1E-4; the initial learning rate is 1E-2, and the momentum parameter is 0.8; Mean Squared Error (MSE) is used as the loss function; the batch size is set to 8, and the total number of training epochs is 200.

[0075] Step 5: Use the exercise power and the first predicted heart rate as inputs to construct the second LSTM model, and output the second predicted heart rate;

[0076] The second LSTM model includes: a first LSTM layer, a Dropout layer, a second LSTM layer, and a fully connected layer in cascade.

[0077] The first LSTM (Long Short-Term Memory) layer of the second LSTM model receives the exercise power data; to optimize the training process of the model, Stochastic Gradient Descent (SGD) is used, and the weight decay is set to 1E-4 to prevent overfitting; the initial learning rate is set to 1E-2, and the momentum parameter is adjusted to 0.8 to accelerate convergence and improve the stability of the model; the design purpose of the first LSTM layer is to capture the temporal features in the input data and generate a context output.

[0078] The second LSTM layer is introduced to process the data after being processed by the first LSTM layer; a bidirectional structure and a dropout strategy are used in the LSTM layer to prevent overfitting and improve the generalization ability of the model.

[0079] During the implementation of the model, Xavier uniform initialization is performed on the weights to ensure the stability and convergence of the model; and the Z-Score normalization technique is applied to normalize the input data to improve the training efficiency and prediction accuracy of the model; finally, the data processed by the stacked LSTM layers will pass through an output layer to generate the final second predicted heart rate result.

[0080] Experimental process:

[0081] This invention uses the CGU dataset of National Taiwan University of Science and Technology. The dataset contains data collected from 75 subjects during stationary cycling and skipping rope exercises; for each exercise form, the subjects collected three groups of data; the data was collected by an IWR6843 radar and transmitted to a computer through a DAC1000 data transmitter for subsequent analysis.

[0082] Table 1 Heart rate prediction results of the CGU dataset

[0083]

[0084] As can be seen from Table 1, the errors between the second predicted heart rate and the actually measured heart rate are relatively small under the two exercises of skipping rope and stationary bicycle, indicating the effectiveness of the present invention.

[0085] Taking the ideal embodiments of the present invention described above as inspiration, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A heart rate detection method based on exercise power, characterized in that: The following steps are involved: Step 1: Collect radar signals under different motion states and corresponding heart rate signals; Step 2: extract the spectrum of the radar signal at different distance values, calculate the volume of the body part using the area of ​​the body part, and obtain the mass of the body part through the volume of the body part; Step 3: Calculate the displacement of body parts; Calculate the acceleration of body parts; calculate physical forces using the acceleration and mass of body parts; calculate the physical workload of body parts based on the physical forces; convert the physical workload into motion power; Step 4: construct a first LSTM model, input the exercise power and the actual heart rate into the first LSTM model to obtain a first predicted heart rate; Step 5: Take the exercise power and the first predicted heart rate as input, build a second LSTM model, and output the second predicted heart rate.

2. The heart rate detection method based on exercise power according to claim 1, characterized in that: The formula for body part mass is: m k =ρμD k C k (4) Where ρ represents mass density, μ is the proportionality coefficient, and C k is the received power of the k-distance bin, D k To receive the power C k Scaling factor for conversion to body part area in k distance bins.

3. The heart rate detection method based on exercise power according to claim 1, characterized in that: The formula for physical workload is: Among them, F k =m k a k , a k is the acceleration and Δd is the displacement of the body part.

4. The heart rate detection method based on exercise power according to claim 3, characterized in that: The formula for body part displacement is: Among them, φ k is the phase difference between two adjacent frames, and λ is the wavelength.

5. The heart rate detection method based on exercise power according to claim 4, characterized in that: The formula for converting physical workload into exercise power is: Among them, T f is the period of radar frame.

6. The heart rate detection method based on exercise power according to claim 1, characterized in that: The first LSTM model consists of a 2-unit LSTM composed of dense layers.

7. The heart rate detection method based on exercise power according to claim 1, characterized in that: The second LSTM model includes: the first LSTM layer, the Dropout layer, the second LSTM layer and the fully connected layer cascaded.

8. The heart rate detection method based on exercise power according to claim 1, characterized in that: Exercise status includes: skipping rope, stationary bicycle, and running on fitness equipment.

9. The heart rate detection method based on exercise power according to claim 1, characterized in that: Radar signals include: FMCW radar.

10. The heart rate detection method based on exercise power according to claim 1, characterized in that: Heart rate signals are collected through wearable devices.

Images