Time domain feature extraction method based on pulse signals
Through sliding window, Gaussian fitting and Fourier series fitting combined with genetic algorithm optimization, the problem of incomplete pulse signal characteristics in the existing technology is solved, and more accurate and physiologically significant pulse signal feature extraction is achieved, supporting more Detailed classification of pulse signal.
Patent Information
- Application Number
- CN202410965068.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-12
AI Technical Summary
The existing pulse signal time domain feature extraction methods are not comprehensive and do not have physiological significance, making it difficult to accurately classify and identify pulse signals.
Using sliding window technology, Gaussian fitting and Fourier series fitting combined with genetic algorithm optimization, the time domain characteristics of pulse signals are extracted, and the number of peaks is obtained through the sliding window, Gaussian fitting and Fourier series fitting are combined. Genetic algorithms optimize parameters until threshold error is reached to obtain more comprehensive time domain features.
It realizes a more comprehensive feature extraction of pulse signals, improves the physiological significance and accuracy of the features, and supports more accurate pulse signal classification and recognition.
Smart Images

Figure CN118924243A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of physiological signal detection, and in particular to a time domain feature extraction method based on a pulse signal. Background Art
[0002] Pulse diagnosis is one of the main methods of TCM in disease diagnosis. It is used to observe the changes in the human pulse. The pulse wave is formed by the heartbeat propagating along the arteries and blood flow. The pulse wave is affected by the interaction of multiple physiological factors such as the elasticity of the blood vessel wall and vascular resistance. It contains rich information and can well reflect the physical state of the human body.
[0003] With the development of computer and signal detection technology, the research on pulse signals has been ongoing. Researchers are committed to converting the subjective judgment of doctors into machine biometric recognition with reference basis, so as to provide unified judgment standards and norms. With the development of machine learning and deep learning, how to achieve more accurate and detailed classification of pulse signals has become a new research topic. Therefore, it is particularly important to extract more comprehensive pulse signal features to provide a basis for the classification and recognition of pulse signals.
[0004] The features extracted by the existing pulse signal time domain feature extraction methods are not comprehensive, and the existing pulse signal time domain feature extraction methods usually use waveform fitting methods, which only focus on the waveform itself, but the extracted features do not have physiological significance. Therefore, it is an urgent problem to provide a method that can extract more comprehensive time domain features of pulse signals, and the extracted time domain features have certain physiological significance. Summary of the invention
[0005] In view of the above analysis, the present invention aims to provide a time domain feature extraction method based on pulse signal, so as to solve the problem that the features extracted by the existing pulse signal time domain feature extraction method are incomplete and the extracted features have no physiological significance.
[0006] The present invention provides a method for extracting time domain features based on a pulse signal, the method comprising the following steps:
[0007] Acquiring a human pulse signal through a sensor, and preprocessing the pulse signal to obtain a preprocessed pulse signal;
[0008] Performing a sliding window on the preprocessed pulse signal to obtain the number of peaks of the pulse signal corresponding to each window, and performing a first group of time domain feature extraction on the preprocessed pulse signal based on the number of peaks;
[0009] Performing period segmentation, quality screening, and period normalization on the preprocessed pulse signal to obtain a period-normalized pulse signal;
[0010] Using a Gaussian function to perform Gaussian fitting on the pulse signal after the period normalization, and extracting a second group of time domain features based on the Gaussian fitting function;
[0011] Fourier series fitting is performed on the pulse signal after the period normalization by using Fourier series, and a third group of time domain features is extracted based on the Fourier series.
[0012] Furthermore, during Gaussian fitting, if the root mean square error of Gaussian fitting does not reach a threshold, the number of Gaussian functions and their parameters are optimized based on a genetic algorithm until the root mean square error of Gaussian fitting reaches the threshold, and a second group of time domain feature extraction is performed based on the Gaussian fitting function corresponding to the root mean square error of Gaussian fitting when the root mean square error of Gaussian fitting reaches the threshold;
[0013] During Fourier series fitting, if the root mean square error of the Fourier series fitting does not reach the threshold, the parameters of the Fourier series are optimized based on the genetic algorithm until the root mean square error of the Fourier series fitting reaches the threshold, and the third group of time domain features are extracted based on the Fourier series corresponding to the root mean square error of the Fourier series fitting when the root mean square error reaches the threshold.
[0014] Furthermore, if the Gaussian fitting root mean square error does not reach the threshold, optimizing the number of Gaussian functions and their parameters based on a genetic algorithm until the Gaussian fitting root mean square error reaches the threshold includes the following steps:
[0015] Step S41, initializing first configuration parameters, generating genetic individuals based on the first configuration parameters to construct a first population;
[0016] Step S42, calculating the Gaussian fitting root mean square error corresponding to each individual in the first population, obtaining the fitness value of each individual according to the Gaussian fitting root mean square error of each individual, and selecting individuals whose fitness values are less than a threshold to form a first excellent population;
[0017] Step S43, determine whether the termination condition is met. If so, obtain the number of Gaussian functions and their parameters based on the individual with the smallest fitness value in the first population; otherwise, add 1 to the number of Gaussian functions of each individual in the first excellent population, and then perform crossover and mutation based on the crossover probability and mutation probability to obtain a new first genetic individual, combine the individuals in the first excellent population with the new first genetic individual to obtain an updated first population, use the updated first population as the population for the next iteration, and repeat steps S42 to S43.
[0018] Furthermore, the current fitness value of the individual is obtained by the following formula:
[0019]
[0020] Among them, ε i is the root mean square error of the Gaussian fitting corresponding to the i-th individual, and n is the number of Gaussian functions.
[0021] Furthermore, if the termination condition is not reached, the number of Gaussian functions in the genes of each individual in the first excellent population is increased by 1 to obtain an updated first excellent population; two individuals in the updated first excellent population are selected as parent individuals by the roulette method according to the crossover probability, and the parameter values of the Gaussian functions of the parent individuals are updated by random multi-point crossover to generate new first genetic individuals; the parameter values of the Gaussian functions of the individuals in the updated first excellent population are randomly changed according to the mutation probability to mutate and generate new first genetic individuals; the individuals in the first excellent population are combined with the new first genetic individuals generated by crossover and mutation to obtain an updated first population.
[0022] Furthermore, if the root mean square error of Fourier series fitting does not reach the threshold, optimizing the parameters of Fourier series based on genetic algorithm until the root mean square error of Fourier series fitting reaches the threshold comprises the following steps:
[0023] Step S51, initializing the second configuration parameters, encoding and generating genetic individuals based on the second configuration parameters to construct a plurality of second populations;
[0024] Step S52, calculating the Fourier series fitting root mean square error corresponding to each individual in each second population after encoding, obtaining the fitness value of each individual according to the Fourier series fitting root mean square error of each individual, and selecting the individual with the smallest fitness value in each second population to form a second excellent population;
[0025] Step S53, determine whether the termination condition is met, if so, obtain the parameters of the Fourier series according to the individual with the smallest fitness value in the current second population; otherwise, crossover and mutation are performed on each individual in the second excellent population according to the crossover probability and the mutation probability to obtain a new second genetic individual, and the individuals in the second excellent population are combined with the new second genetic individuals to obtain multiple updated second populations, and the updated second population is used as the population for the next iteration, and steps S52 to S53 are repeated.
[0026] Furthermore, if the termination condition is not reached, two individuals in the second excellent population are selected as parent individuals by the roulette method according to the crossover probability, and new second genetic individuals are generated by random multi-point crossover of the parent individuals; the values of the Fourier series parameters of the individuals in the second excellent population are randomly changed according to the mutation probability to mutate and generate new second genetic individuals; the individuals in the second excellent population are combined with the new second genetic individuals generated by crossover and mutation to obtain multiple updated second populations.
[0027] Furthermore, the step of performing period segmentation, quality screening, and period normalization on the preprocessed pulse signal to obtain a period-normalized pulse signal includes:
[0028] Step S31, performing period segmentation on the preprocessed pulse signal based on the pacemaker point;
[0029] Step S32, calculating the peak-to-valley slope, peak-to-valley distance, and peak-to-valley difference of the pulse signal of each segmented period, obtaining the arithmetic mean of the peak-to-valley slope and the arithmetic mean of the peak-to-valley distance based on the peak-to-valley slope and the peak-to-valley distance of all segmented pulse signals, respectively, removing the periodic pulse signals whose peak-to-valley slope deviates from the peak-to-valley slope average value by more than a threshold or whose peak-to-valley distance deviates from the peak-to-valley distance average value by more than a threshold from each periodic signal, and obtaining the remaining periodic pulse signals;
[0030] Step S33, obtaining the average cycle length of each remaining periodic pulse signal; stretching or compressing each remaining periodic pulse signal to normalize the cycle.
[0031] Further, the preprocessing the pulse signal to obtain a preprocessed pulse signal includes:
[0032] Setting a noise threshold, determining a wavelet basis function and a decomposition layer number to perform wavelet decomposition on the pulse signal to obtain a plurality of wavelet components, denoising the wavelet components according to the noise threshold, and then reconstructing the signal using wavelet coefficients to obtain a denoised pulse signal;
[0033] The pacing points of the denoised pulse signal are extracted by searching for local extreme values, and the baseline noise is fitted by cubic spline interpolation on all the pacing points. The baseline noise is removed from the denoised pulse signal to obtain the preprocessed pulse signal.
[0034] Furthermore, the size of the sliding window and the moving step are set, and local extreme points of the pulse signal corresponding to the sliding window after each sliding are extracted based on the sliding window and the moving step; a first-order difference is performed on the pulse signal corresponding to the sliding window after each sliding to obtain the number of zero points of the first-order difference signal, and the number of peaks of the corresponding pulse signal in the sliding window is obtained based on the number of zero points; and a first group of time domain features are extracted from the preprocessed pulse signal based on the number of peaks and local extreme points.
[0035] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0036] 1. The present invention extracts 54 time domain features of the pulse signal through sliding window, Gaussian fitting and Fourier series fitting, thereby realizing a more comprehensive time domain feature extraction of the pulse signal and providing a basis for more accurate and detailed classification and identification of the pulse signal.
[0037] 2. The present invention matches the features with physiological meanings when extracting features through a sliding window, thereby solving the problem that the extracted features do not have physiological meanings.
[0038] 3. The present invention optimizes the number of Gaussian functions and their parameters by genetic algorithm during Gaussian fitting, thereby improving the fitting accuracy, so that the extracted features can more accurately reflect the pulse information.
[0039] 4. The present invention optimizes the parameters of the Fourier series by means of a genetic algorithm during Fourier series fitting, thereby improving the fitting accuracy, so that the extracted features can more accurately reflect the pulse information.
[0040] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.
[0042] Figure 1 Flow chart of a pulse signal time domain feature extraction method according to an embodiment of the present invention;
[0043] Figure 2 Schematic diagram of the amplitude-frequency comparison of the pulse signal after wavelet transform denoising according to an embodiment of the present invention;
[0044] Figure 3 This is a schematic diagram of cubic spline interpolation fitting according to an embodiment of the present invention;
[0045] Figure 4 A schematic diagram of a pulse signal waveform obtained by a sliding window according to an embodiment of the present invention;
[0046] Figure 5 A schematic diagram of Gaussian fitting of a pulse signal according to an embodiment of the present invention;
[0047] Figure 6 Schematic diagram of Fourier series fitting of a pulse signal according to an embodiment of the present invention. DETAILED DESCRIPTION
[0048] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0049] A specific embodiment of the present invention discloses a method for extracting time domain features based on pulse signals. Figure 1 As shown, the method comprises the following steps:
[0050] Step S1, obtaining a human pulse signal through a sensor, and preprocessing the pulse signal to obtain a preprocessed pulse signal;
[0051] Step S2, sliding a window on the preprocessed pulse signal to obtain the number of peaks of the pulse signal corresponding to each window, and extracting a first group of time domain features from the preprocessed pulse signal based on the number of peaks;
[0052] Step S3, performing period segmentation, quality screening, and period normalization on the pre-processed pulse signal to obtain a period-normalized pulse signal;
[0053] Step S4, using a Gaussian function to perform Gaussian fitting on the pulse signal after the period normalization, and extracting a second group of time domain features based on the Gaussian fitting function;
[0054] Step S5, using Fourier series to perform Fourier series fitting on the pulse signal after the period normalization, and extracting a third group of time domain features based on the Fourier series.
[0055] Specifically, in step S1, preprocessing the pulse signal to obtain a preprocessed pulse signal includes:
[0056] Setting a noise threshold, determining a wavelet basis function and a decomposition layer number to perform wavelet decomposition on the pulse signal to obtain a plurality of wavelet components, denoising the wavelet components according to the noise threshold, and then reconstructing the signal using wavelet coefficients to obtain a denoised pulse signal;
[0057] The pacing points of the denoised pulse signal are extracted by searching for local extreme points, and the baseline noise is fitted by cubic spline interpolation on all the pacing points. The baseline noise is removed from the denoised pulse signal to obtain the preprocessed pulse signal.
[0058] Specifically, the pulse signal is the trajectory of vascular pulsation and is a periodic signal. The pulse signal contains a main wave, a tidal wave (pre-dicrotic wave) and a dicrotic wave in each cycle. The main wave amplitude reflects left ventricular ejection and compliance of the aorta, the tidal wave amplitude reflects vascular elasticity and peripheral resistance, and the dicrotic wave amplitude reflects the elasticity of the aorta and the state of the aortic valve. The time domain characteristics of the pulse signal can reflect the change of its frequency over time and the instantaneous frequency and amplitude corresponding to each moment. The human body's pulse signal is sampled by a sensor with a sampling frequency of 125Hz, and a total of 10,000 sampling data are obtained.
[0059] Furthermore, a low-frequency noise threshold and a high-frequency noise threshold are set, and the pulse signal is subjected to a 7-layer wavelet decomposition using the sym8 wavelet basis function to remove the wavelet components with frequencies lower than the low-frequency noise threshold and higher than the high-frequency noise threshold. The remaining wavelet components are reconstructed using wavelet coefficients to obtain a denoised pulse signal, such as Figure 2 shown.
[0060] Preferably, setting 0.825 Hz as the low-frequency noise threshold and setting 31.25 Hz as the high-frequency noise threshold can remove low-frequency noise generated by skin friction, 50 Hz power frequency interference, and 30-120 Hz high-frequency noise of ECG signals, thereby improving the signal-to-noise ratio.
[0061] Furthermore, if Figure 3 As shown, a search step is set, and the local minimum of the denoised pulse signal is searched within the search step range as the pacing point. All the pacing points are used as interpolation points and cubic spline interpolation fitting is performed on them to obtain the baseline noise. The baseline noise is subtracted from the denoised pulse signal to obtain the preprocessed pulse signal.
[0062] Preferably, since the pulse signal period is usually between 80 and 110, the search step length is set to 50.
[0063] It can be understood that the pulse signal contains not only high and low frequency noise, but also baseline drift caused by breathing. By using cubic spline interpolation to curve fit the pacemaker point to simulate breathing and remove it, a smooth pulse signal with a high signal-to-noise ratio can be obtained.
[0064] Specifically, in step S2, the size of the sliding window and the moving step are set, and the local extreme points of the pulse signal corresponding to the sliding window after each sliding are extracted based on the sliding window and the moving step; a first-order difference is performed on the pulse signal corresponding to the sliding window after each sliding to obtain the number of zero points of the first-order difference signal, and the number of peaks of the corresponding pulse signal in the sliding window is obtained based on the number of zero points; and a first group of time domain features are extracted from the preprocessed pulse signal based on the number of peaks and the local extreme points.
[0065] Specifically, the distance between adjacent pacing points is obtained as the size of the sliding window and the moving step length, the sliding window is slid from the first pacing point according to the moving step length, and the local maximum and local minimum points of the corresponding pulse signal in the sliding window after each sliding are extracted.
[0066] Specifically, the difference between the horizontal coordinates of adjacent pacemaker points is used as the distance between them. After each sliding, there may be multiple local maximum points corresponding to the pulse signal in the window. Similarly, there may be multiple local minimum points corresponding to the pulse signal in the window after each sliding.
[0067] Furthermore, after each sliding, a first-order difference is performed on the corresponding pulse signal in the sliding window to obtain the number of zero points of the first-order difference signal, and the number of peaks of the corresponding pulse signal in the sliding window is obtained based on the number of zero points.
[0068] Specifically, the pulse signal waveform obtained by the sliding window is as follows: Figure 4 As shown, the difference between the adjacent sampling point values of the pulse signal corresponding to the sliding window is taken to obtain a first-order differential signal, and the point where the sign changes in the first-order differential signal is found to obtain the zero point, and the number of zero points is counted. If the number of zero points is greater than or equal to 1 and less than 3, the number of peaks of the corresponding pulse signal in the sliding window is 1, that is, a single-peak signal; if the number of zero points is greater than or equal to 3 and less than or equal to 4, the number of peaks of the corresponding pulse signal in the sliding window is 2, that is, a double-peak signal; if the number of zero points is greater than 4, the number of peaks of the corresponding pulse signal in the sliding window is 3, that is, a triple-peak signal.
[0069] It can be understood that in a discrete signal, a first-order differential signal can be obtained by taking the difference between the data values of adjacent sampling points. The present application adopts backward difference, that is, the data value of the next sampling point is subtracted from the data value of the current sampling point to obtain the first-order differential signal of the current sampling point. The sign change refers to the sign of the differential signal data value changing from positive to negative, or from negative to positive from one sampling point to the next sampling point, so the point where the sign changes is recorded as the zero point. According to the characteristics of the pulse signal, it can be known from the simulation experiment that the corresponding pulse signal in the sliding window is composed of one, two or three peaks, so when the number of zero points of the first-order differential signal is greater than or equal to 1 and less than 3, the number of peaks of the corresponding pulse signal in the sliding window is 1, and when the number of zero points is greater than or equal to 3 and less than or equal to 4, the number of peaks of the corresponding pulse signal is 2, and when the number of zero points is greater than 4, the number of peaks of the corresponding pulse signal is 3.
[0070] Furthermore, if the number of peaks is 1, the first group of time domain features as shown in Table 1 is extracted from the preprocessed pulse signal.
[0071] Table 1 The first set of time domain features extracted when the peak of the pulse signal in the sliding window is 1
[0072]
[0073] Furthermore, if the number of the peaks is 2, the first group of time domain features as shown in Table 2 is extracted from the preprocessed pulse signal.
[0074] Table 2 The first set of time domain features extracted when the peak of the pulse signal in the sliding window is 2
[0075]
[0076]
[0077] Furthermore, if the number of the peaks is 3, the first group of time domain features as shown in Table 3 is extracted from the preprocessed pulse signal.
[0078] Table 3 The first set of time domain features extracted when the peak of the pulse signal in the sliding window is 3
[0079]
[0080]
[0081] Specifically, in step S3, the step of performing period segmentation, quality screening, and period normalization on the preprocessed pulse signal to obtain a period-normalized pulse signal includes:
[0082] Step S31, performing period segmentation on the preprocessed pulse signal based on the pacemaker point;
[0083] Step S32, calculating the peak-to-valley slope, peak-to-valley distance, and peak-to-valley difference of the pulse signal of each segmented period, obtaining the arithmetic mean of the peak-to-valley slope and the arithmetic mean of the peak-to-valley distance based on the peak-to-valley slope and the peak-to-valley distance of all segmented pulse signals, respectively, removing the periodic pulse signals whose peak-to-valley slope deviates from the peak-to-valley slope average value by more than a threshold or whose peak-to-valley distance deviates from the peak-to-valley distance average value by more than a threshold from each periodic signal, and obtaining the remaining periodic pulse signals;
[0084] Step S33, obtaining the average cycle length of each remaining periodic pulse signal; stretching or compressing each remaining periodic pulse signal to normalize the cycle.
[0085] Specifically, in step S31, the signal between adjacent pacemakers is a periodic signal, and the preprocessed pulse signal is divided into periods according to the pacemakers.
[0086] Specifically, in step S32, the peak-to-valley slope, peak-to-valley distance, and peak-to-valley difference are calculated for each segmented periodic signal, where the peak-to-valley slope is the slope of the line connecting the peak point and the pacemaker point of the periodic signal, that is, the ratio of the difference between the vertical coordinates of the peak point and the pacemaker point to the difference between the horizontal coordinates of the peak point and the pacemaker point; the peak-to-valley distance is the difference between the horizontal coordinates of the peak point and the pacemaker point of the periodic signal, and the peak-to-valley difference is the difference between the vertical coordinates of the peak point and the pacemaker point of the periodic signal.
[0087] Preferably, the periodic pulse signals whose peak-to-valley slope deviates from the average value of the peak-to-valley slope by more than 10% or whose peak-to-valley distance deviates from the average value of the peak-to-valley distance by more than 10% are removed.
[0088] Specifically, in step S33, for each periodic pulse signal after removal, if the length of the signal is greater than the average cycle length, the signal is truncated according to the average cycle length; if the length of the signal is less than the average cycle length, the signal is padded with zeros according to the average cycle length.
[0089] It can be understood that in the process of processing a large number of different pulse signals, amplitude normalization can eliminate the dimensional data differences of the pulse signal and retain the shape of the signal's main peak, pre-dicrotic wave, and dicrotic wave; period normalization retains the relative positions of characteristic points such as the signal's main peak, pre-dicrotic wave, and dicrotic wave, which facilitates subsequent analysis and processing.
[0090] Specifically, in step S4, during Gaussian fitting, if the root mean square error of the Gaussian fitting does not reach the threshold, the number of Gaussian functions and their parameters are optimized based on the genetic algorithm until the root mean square error of the Gaussian fitting reaches the threshold, and the second group of time domain features are extracted based on the Gaussian fitting function corresponding to the root mean square error of the Gaussian fitting when the root mean square error of the Gaussian fitting reaches the threshold.
[0091] Specifically, Gaussian fitting is performed by the following method:
[0092]
[0093] Among them, t is the independent variable, A i , B i , C i is the parameter of the i-th Gaussian function, and n is the number of Gaussian functions.
[0094] Furthermore, if the Gaussian fitting root mean square error does not reach the threshold, optimizing the number of Gaussian functions and their parameters based on a genetic algorithm until the Gaussian fitting root mean square error reaches the threshold includes the following steps:
[0095] Step S41, initializing first configuration parameters, generating genetic individuals based on the first configuration parameters to construct a first population;
[0096] Step S42, calculating the Gaussian fitting root mean square error corresponding to each individual in the first population, obtaining the fitness value of each individual according to the Gaussian fitting root mean square error of each individual, and selecting individuals whose fitness values are less than a threshold to form a first excellent population;
[0097] Step S43, determine whether the termination condition is met. If so, obtain the number of Gaussian functions and their parameters based on the individual with the smallest fitness value in the first population; otherwise, add 1 to the number of Gaussian functions of each individual in the first excellent population, and then perform crossover and mutation based on the crossover probability and mutation probability to obtain a new first genetic individual, combine the individuals in the first excellent population with the new first genetic individual to obtain an updated first population, use the updated first population as the population for the next iteration, and repeat steps S42 to S43.
[0098] Specifically, in step S41, the first configuration parameters include the number of individuals in the population, the number of Gaussian functions, and the Gaussian function parameter A. i , B i , C i The value range, crossover probability and mutation probability of . Genetic individuals are generated by encoding based on the number of Gaussian functions and their parameters, expressed as X = [n, A1, B1, C1...A n ,B n ,C n ], where n is the number of Gaussian functions, and A1, B1, and C1 are the parameters of the first Gaussian function. i , B i , C i Randomly select numerical values within the value range of to generate genetic individuals to complete the construction of multiple first initial populations.
[0099] For example, the maximum population size is set to 1000, the number of individuals in the population is 400, the number of initial Gaussian functions is 1, the crossover rate is 0.8, the mutation rate is 0.25, and A i The value range is [0,3], B i The value range is [0,85], C i The value range is [0,80].
[0100] Specifically, in step S42, n corresponding Gaussian functions are obtained according to the specific values of each individual in each group after encoding, and the Gaussian fitting root mean square error when the n Gaussian functions are fitted to the pulse signal after period normalization is calculated.
[0101] Furthermore, the current fitness value of the individual is obtained by the following formula:
[0102]
[0103] Among them, ε i is the root mean square error of the Gaussian fitting corresponding to the i-th individual, and n is the number of Gaussian functions.
[0104] Specifically, in step S43, the termination condition includes that the Gaussian fitting root mean square error of the individual with the smallest fitness value in the current first population is less than a threshold or the number of iterations reaches a maximum value.
[0105] Preferably, the Gaussian fitting root mean square error is less than 7.5*10 -3 , the maximum number of iterations is 400.
[0106] Furthermore, if the termination condition is not reached, the number of Gaussian functions in the genes of each individual in the first excellent population is increased by 1 to obtain an updated first excellent population; two individuals in the updated first excellent population are selected as parent individuals by the roulette method according to the crossover probability, and the parameter values of the Gaussian functions of the parent individuals are updated by random multi-point crossover to generate new first genetic individuals; the parameter values of the Gaussian functions of the individuals in the updated first excellent population are randomly changed according to the mutation probability to mutate and generate new first genetic individuals; the individuals in the first excellent population are combined with the new first genetic individuals generated by crossover and mutation to obtain an updated first population.
[0107] Furthermore, the pulse signal Gaussian fitting is as follows: Figure 5 As shown, the parameters of all Gaussian fitting functions corresponding to when the Gaussian fitting root mean square error reaches a threshold are obtained as the second group of time domain features.
[0108] Specifically, when the Gaussian fitting root mean square error reaches the threshold, the number of Gaussian functions is 4, and the parameters of each Gaussian function are 3. Therefore, the second group of time domain features extracted by Gaussian function fitting is 12. The preprocessed pulse signal can be restored by the second group of time domain features.
[0109] Specifically, in step S5, when Fourier series fitting, if the root mean square error of Fourier series fitting does not reach the threshold, the parameters of the Fourier series are optimized based on the genetic algorithm until the root mean square error of Fourier series fitting reaches the threshold, and the third group of time domain features are extracted based on the Fourier series corresponding to the root mean square error of Fourier series fitting when the root mean square error reaches the threshold.
[0110] Specifically, Fourier series fitting is performed by the following method:
[0111]
[0112] Among them, a0 is a constant term; a i is the coefficient that determines the amplitude of the cosine function; b i is the coefficient, which determines the amplitude of the sine function; w is the fundamental frequency; n is the number of harmonics, n=7.
[0113] Furthermore, if the root mean square error of Fourier series fitting does not reach the threshold, optimizing the parameters of Fourier series based on genetic algorithm until the root mean square error of Fourier series fitting reaches the threshold comprises the following steps:
[0114] Step S51, initializing the second configuration parameters, encoding and generating genetic individuals based on the second configuration parameters to construct a plurality of second populations;
[0115] Step S52, calculating the Fourier series fitting root mean square error corresponding to each individual in each second population after encoding, obtaining the fitness value of each individual according to the Fourier series fitting root mean square error of each individual, and selecting the individual with the smallest fitness value in each second population to form a second excellent population;
[0116] Step S53, determine whether the termination condition is met, if so, obtain the parameters of the Fourier series according to the individual with the smallest fitness value in the current second population; otherwise, crossover and mutation are performed on each individual in the second excellent population according to the crossover probability and the mutation probability to obtain a new second genetic individual, and the individuals in the second excellent population are combined with the new second genetic individuals to obtain multiple updated second populations, and the updated second population is used as the population for the next iteration, and steps S52 to S53 are repeated.
[0117] Specifically, in step S51, the second configuration parameters include the maximum population size, the number of individuals in the population, the Fourier series parameter a i , b i The value range, crossover probability and mutation probability of . Genetic individuals are generated by encoding based on the parameters of the Fourier series, expressed as X = [a0, a1, b1, ... a7, b7]. i , b i Randomly select values within the value range to generate genetic individuals to complete the construction of multiple second initial populations.
[0118] For example, the maximum population size is set to 500, the number of individuals in the population is set to 200, the order of the Fourier series is set to 7, the crossover rate is set to 0.5, the mutation rate is set to 0.2, and a i , b i The value range is [-2,2].
[0119] Specifically, in step S52, the Fourier series corresponding to each individual in each population is obtained according to the specific value after encoding, and the Fourier series fitting root mean square error when the Fourier series is fitted with the pulse signal after period normalization is calculated.
[0120] Furthermore, the current fitness value of the individual is obtained by the following formula:
[0121]
[0122] Among them, RMSE i is the root mean square error of the Fourier series fitting corresponding to the i-th individual.
[0123] Specifically, in step S53, the termination condition includes that the Gaussian fitting root mean square error of the individual with the smallest fitness value in the current second population is less than a threshold or the number of iterations reaches a maximum value.
[0124] Preferably, the root mean square error of Fourier series fitting is less than 2.5*10 -3 , the maximum number of iterations is 100.
[0125] Specifically, if the termination condition is not reached, two individuals in the second best population are selected as parent individuals by the roulette method according to the crossover probability, and new second genetic individuals are generated by random multi-point crossover of the parent individuals; the values of the Fourier series parameters of the individuals in the second best population are randomly changed according to the mutation probability to mutate and generate new second genetic individuals; the individuals in the second best population are combined with the new second genetic individuals generated by crossover and mutation to obtain multiple updated second populations.
[0126] Furthermore, the pulse signal Fourier series fitting is as follows: Figure 6 As shown, the parameters of all Fourier series corresponding to when the root mean square error of Fourier series fitting reaches a threshold are obtained as the third group of time domain features.
[0127] Specifically, when the root mean square error of Fourier series fitting reaches the threshold, the order of the Fourier series is 7, and the parameters of each Fourier series are 2, plus the constant term, so the third group of time domain features extracted by Fourier series fitting is 15. The preprocessed pulse signal can be restored by the third group of time domain features.
[0128] Compared with the prior art, the pulse signal time domain feature extraction method provided by the present invention has the following beneficial effects:
[0129] 1. The present invention extracts 54 time domain features of the pulse signal through sliding window, Gaussian fitting and Fourier series fitting, thereby realizing a more comprehensive time domain feature extraction of the pulse signal and providing a basis for more accurate and detailed classification and identification of the pulse signal.
[0130] 2. The present invention matches the features with physiological meanings when extracting features through a sliding window, thereby solving the problem that the extracted features do not have physiological meanings.
[0131] 3. The present invention optimizes the number of Gaussian functions and their parameters by genetic algorithm during Gaussian fitting, thereby improving the fitting accuracy, so that the extracted features can more accurately reflect the pulse information.
[0132] 4. The present invention optimizes the parameters of the Fourier series by means of a genetic algorithm during Fourier series fitting, thereby improving the fitting accuracy, so that the extracted features can more accurately reflect the pulse information.
[0133] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.
[0134] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A time domain feature extraction method based on pulse signal, characterized in that: The method comprises the following steps: Acquiring a human pulse signal through a sensor, and preprocessing the pulse signal to obtain a preprocessed pulse signal; Performing a sliding window on the preprocessed pulse signal to obtain the number of peaks of the pulse signal corresponding to each window, and performing a first group of time domain feature extraction on the preprocessed pulse signal based on the number of peaks; Performing period segmentation, quality screening, and period normalization on the preprocessed pulse signal to obtain a period-normalized pulse signal; Using a Gaussian function to perform Gaussian fitting on the pulse signal after the period normalization, and extracting a second group of time domain features based on the Gaussian fitting function; Fourier series fitting is performed on the pulse signal after the period normalization by using Fourier series, and a third group of time domain features is extracted based on the Fourier series.
2. The method for extracting time domain features based on pulse signals according to claim 1, characterized in that: During Gaussian fitting, if the root mean square error of Gaussian fitting does not reach the threshold, the number of Gaussian functions and their parameters are optimized based on the genetic algorithm until the root mean square error of Gaussian fitting reaches the threshold, and the second group of time domain features are extracted based on the Gaussian fitting function corresponding to the root mean square error of Gaussian fitting when the root mean square error of Gaussian fitting reaches the threshold; During Fourier series fitting, if the root mean square error of the Fourier series fitting does not reach the threshold, the parameters of the Fourier series are optimized based on the genetic algorithm until the root mean square error of the Fourier series fitting reaches the threshold, and the third group of time domain features are extracted based on the Fourier series corresponding to the root mean square error of the Fourier series fitting when the root mean square error reaches the threshold.
3. The method for extracting time domain features based on pulse signals according to claim 2, characterized in that: If the Gaussian fitting root mean square error does not reach the threshold, optimizing the number of Gaussian functions and their parameters based on a genetic algorithm until the Gaussian fitting root mean square error reaches the threshold includes the following steps: Step S41, initializing first configuration parameters, generating genetic individuals based on the first configuration parameters to construct a first population; Step S42, calculating the Gaussian fitting root mean square error corresponding to each individual in the first population, obtaining the fitness value of each individual according to the Gaussian fitting root mean square error of each individual, and selecting individuals whose fitness values are less than a threshold to form a first excellent population; Step S43, determine whether the termination condition is met. If so, obtain the number of Gaussian functions and their parameters based on the individual with the smallest fitness value in the first population; otherwise, add 1 to the number of Gaussian functions of each individual in the first excellent population, and then perform crossover and mutation based on the crossover probability and mutation probability to obtain a new first genetic individual, combine the individuals in the first excellent population with the new first genetic individual to obtain an updated first population, use the updated first population as the population for the next iteration, and repeat steps S42 to S43.
4. The method for extracting time domain features based on pulse signals according to claim 3, characterized in that: The current fitness value of the individual is obtained by the following formula: Among them, ε i is the root mean square error of the Gaussian fitting corresponding to the i-th individual, and n is the number of Gaussian functions.
5. The method for extracting time domain features based on pulse signals according to claim 3, characterized in that: If the termination condition is not reached, the number of Gaussian functions in the gene of each individual in the first excellent population is increased by 1 to obtain an updated first excellent population; two individuals in the updated first excellent population are selected as parent individuals by the roulette method according to the crossover probability, and the parameter values of the Gaussian functions of the parent individuals are updated by random multi-point crossover to generate new first genetic individuals; the parameter values of the Gaussian functions of the individuals in the updated first excellent population are randomly changed according to the mutation probability to mutate and generate new first genetic individuals; the individuals in the first excellent population are combined with the new first genetic individuals generated by crossover and mutation to obtain an updated first population.
6. The method for extracting time domain features based on pulse signals according to claim 2, characterized in that: If the root mean square error of Fourier series fitting does not reach the threshold, optimizing the parameters of Fourier series based on genetic algorithm until the root mean square error of Fourier series fitting reaches the threshold comprises the following steps: Step S51, initializing the second configuration parameters, encoding and generating genetic individuals based on the second configuration parameters to construct a plurality of second populations; Step S52, calculating the Fourier series fitting root mean square error corresponding to each individual in each second population after encoding, obtaining the fitness value of each individual according to the Fourier series fitting root mean square error of each individual, and selecting the individual with the smallest fitness value in each second population to form a second excellent population; Step S53, determine whether the termination condition is met, if so, obtain the parameters of the Fourier series according to the individual with the smallest fitness value in the current second population; otherwise, crossover and mutation are performed on each individual in the second excellent population according to the crossover probability and the mutation probability to obtain a new second genetic individual, and the individuals in the second excellent population are combined with the new second genetic individuals to obtain multiple updated second populations, and the updated second population is used as the population for the next iteration, and steps S52 to S53 are repeated.
7. The method for extracting time domain features based on pulse signals according to claim 6, characterized in that: If the termination condition is not met, two individuals in the second best population are selected as parent individuals by the roulette method according to the crossover probability, and new second genetic individuals are generated by random multi-point crossover of the parent individuals; the values of the Fourier series parameters of the individuals in the second best population are randomly changed according to the mutation probability to mutate and generate new second genetic individuals; the individuals in the second best population are combined with the new second genetic individuals generated by crossover and mutation to obtain multiple updated second populations.
8. The method for extracting time domain features based on pulse signals according to claim 1, characterized in that: The step of performing period segmentation, quality screening, and period normalization on the pre-processed pulse signal to obtain a period-normalized pulse signal includes: Step S31, performing period segmentation on the preprocessed pulse signal based on the pacemaker point; Step S32, calculating the peak-to-valley slope, peak-to-valley distance, and peak-to-valley difference of the pulse signal of each segmented period, obtaining the arithmetic mean of the peak-to-valley slope and the arithmetic mean of the peak-to-valley distance based on the peak-to-valley slope and the peak-to-valley distance of all segmented pulse signals, respectively, removing the periodic pulse signals whose peak-to-valley slope deviates from the peak-to-valley slope average value by more than a threshold or whose peak-to-valley distance deviates from the peak-to-valley distance average value by more than a threshold from each periodic signal, and obtaining the remaining periodic pulse signals; Step S33, obtaining the average cycle length of each remaining periodic pulse signal; stretching or compressing each remaining periodic pulse signal to normalize the cycle.
9. The method for extracting time domain features based on pulse signals according to claim 1, characterized in that: The preprocessing of the pulse signal to obtain a preprocessed pulse signal comprises: Setting a noise threshold, determining a wavelet basis function and a decomposition layer number to perform wavelet decomposition on the pulse signal to obtain a plurality of wavelet components, denoising the wavelet components according to the noise threshold, and then reconstructing the signal using wavelet coefficients to obtain a denoised pulse signal; The pacing points of the denoised pulse signal are extracted by searching for local extreme values, and the baseline noise is fitted by cubic spline interpolation on all the pacing points. The baseline noise is removed from the denoised pulse signal to obtain the preprocessed pulse signal.
10. The method for extracting time domain features based on pulse signals according to claim 1, characterized in that: The size of the sliding window and the moving step length are set, and the local extreme value points of the pulse signal corresponding to the sliding window after each sliding are extracted based on the sliding window and the moving step length; the first-order difference of the pulse signal corresponding to the sliding window after each sliding is performed to obtain the number of zero points of the first-order difference signal, and the number of peaks of the corresponding pulse signal in the sliding window is obtained based on the number of zero points; A first set of time domain features is extracted from the preprocessed pulse signal based on the number of wave peaks and local extreme points.
Citation Information
Patent Citations
Pulse feature extraction device and method
CN105433916A
Pulse wave signal denoising method based on DTCWT-Spline
CN107184187A
Biological information measurement device, biological information measurement program, and biological information measurement method
CN111386071A
Method for screening desynchronization index
CN112423648A
Noninvasive blood glucose detection method based on time-frequency domain comprehensive analysis
CN113662538A
Cited By
Pulse wave signal time domain feature extraction method based on flexible sensor
CN121176868A
