A pulse wave time domain feature classification system, identification system and computer readable storage medium

Abstract

The present application belongs to the field of pulse wave time domain feature recognition, and particularly relates to a pulse wave time domain feature classification system, a recognition system and a computer readable storage medium. The system of the present application comprises: an input module configured to input a pulse wave; a prediction module configured to identify the pulse wave through a machine learning model, classify the pre-systolic wave and the systolic wave into three features of extreme point, convex-concave point with single-point decreasing and feature fusion respectively, and classify the pulse wave into nine waveform types of A-type pulse wave to I-type pulse wave according to the calculation result; and an output module configured to output the classification result of the prediction module. The pulse wave time domain feature parameters obtained by the present application improve the accuracy of pulse wave time domain feature extraction and the classification precision, and can be used to establish a pulse or disease recognition model, which has a guiding role for the diagnosis of various diseases in Tibetan medicine.

Description

Technical Field

[0001] This invention belongs to the field of pulse wave time-domain feature recognition, specifically relating to a pulse wave time-domain feature classification system, recognition system, and computer-readable storage medium. Background Technology

[0002] Pulse diagnosis is an important component of the four diagnostic methods in Tibetan medicine. The development of a Tibetan pulse wave acquisition device is of great significance for the objective quantification of Tibetan pulse diagnosis. The temporal characteristics of a single-cycle pulse wave are of great importance for determining Tibetan pulse patterns, identifying constitutions such as the Tibetan Dragon type, Tripa, and Pekan, and for modeling diseases.

[0003] A complete pulse wave mainly consists of the main peak, the dicrotic wave, and the dicrotic wave precursor. For example... Figure 1 As shown, the main wave peak is point c, the trough of the dicrotic pre-wave is point d, the peak of the dicrotic pre-wave is point e, the trough of the dicrotic wave is point f, and the peak of the dicrotic wave is point g. When blood vessels have good elasticity and blood flow in the body is stable, a complete pulse wave cycle generally includes three main characteristic waves: the main wave, the dicrotic pre-wave, and the dicrotic wave. Due to certain reasons (such as high fever, pregnancy, etc.), the increase in blood flow within the cross-sectional area of ​​the blood vessel per unit time enhances the local expansion and contraction capacity of the human arteries. The pulse wave shows that the dicrotic pre-wave is close to the dicrotic wave; when the blood vessels dilate further, the dicrotic pre-wave will merge with the dicrotic wave. Conversely, due to certain reasons (such as colds, high blood pressure, arteriosclerosis, etc.), the stiffness of the human blood vessel walls increases. On the pulse wave, the dicrotic pre-wave is close to the main wave peak; when the stiffness does not increase, the dicrotic pre-wave merges with the main wave peak, or even the position of the dicrotic pre-wave exceeds the main wave peak, as shown. Therefore, classifying pulse waves according to their morphological characteristics has a guiding role in the diagnosis of various diseases in Tibetan medicine.

[0004] Currently, the main methods for extracting time-domain feature parameters of pulse waves include: slope method, slope and threshold combination method, differential method, extremum method, wavelet method, and sentence method. The slope method primarily identifies feature pulse waves by finding points of slope change. Early methods relied on these feature points, but this approach is less effective for waveforms with insignificant slope changes or significant interference. The slope and threshold combination method, building upon the slope method, determines the location of feature points by fixing the search range for specific features. However, this algorithm can only identify feature points for a certain type of pulse wave, as its recognition performance is less than ideal when the pulse wave is relatively flat. The differential method uses the spline function to obtain the differential plot of a single-cycle pulse wave. A threshold is set, and the position of the first zero-crossing point after differentiation is designated as point c, with the remaining points designated as d, e, f, and g. This method is only applicable when both the dicrotic wave and the pre-dicrotic wave have peaks, and the value of the first zero-crossing point is not necessarily point c. The extremum method finds the maximum value in the pulse wave, designating the first maximum as point c, and then finding the second and third largest extrema as points e and g, respectively. Like the differential method, this method ignores the fact that the maximum is not necessarily point c, and has a larger error when the dicrotic wave and pre-dicrotic wave are not obvious. The wavelet transform method mainly uses wavelet transform (e.g., Gaussian wavelet) to highlight the time-domain feature points of the pulse wave and extract these feature points. However, this method has poor recognition performance for points d and e. The syntactic pattern recognition method mainly involves classifying the pulse wave signal and then extracting feature parameters from the classified pulse wave signals.

[0005] While time-domain feature parameter extraction methods may perform well in certain specific situations, their limitations are significant in applications requiring complex signals and high precision. Therefore, a new classification standard for single-cycle pulse waves is needed to enable physicians to diagnose diseases more accurately based on the morphological characteristics of the pulse wave. Summary of the Invention

[0006] To address the problems of existing technologies, this invention provides a pulse wave time-domain feature classification system, an identification system, and a computer-readable storage medium.

[0007] A pulse wave time-domain feature classification system includes: The sampling module is configured to collect sampling point data, with a threshold range of 1-0.7N for the sampling points, where N is the total number of sampling points. The classification module is configured to identify the morphological features of the dicrotic pre-wave and dicrotic wave in the pulse wave within a threshold range of the sampling points, and classify the pulse wave into Class A to Class I pulse waves based on the morphological features. The morphological characteristics of pulse waves from Class A to Class I are as follows: (1) Within the threshold range of the sampling point, there are three peaks in the pulse wave. The first peak is the main peak, the second peak is the peak of the diabetic pre-wave, and the valley to its left is the valley of the diabetic pre-wave. The third peak is the peak of the diabetic wave, and the valley to its left is the valley of the diabetic wave. The diabetic pre-wave and diabetic wave are marked as extreme points, and the pulse wave is classified as a type A pulse wave. (2) Within the threshold range of the sampling point, the pulse wave has two peaks from left to right and a single point decreasing convex and concave point. The first peak is the main wave peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point. The single point decreasing convex and concave point is the diabetic wave. The diabetic wave is marked as the single point decreasing convex and concave point. The pulse wave is classified as a type B pulse wave. (3) Within the threshold range of the sampling points, there are two peaks in the pulse wave from left to right; the two points corresponding to the decrease of 1 / 3 of the peak value of the main wave are a and b. The difference between the sampling point coordinate value corresponding to point b and the sampling point coordinate value corresponding to point a is the pulse width w, and the ratio of the pulse width to the total number of sampling points N is w_t. If w_t is greater than the threshold, the features of the diabetic wave and the diabetic pre-wave are fused. The first peak is the main peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point, the diabetic wave is marked as the feature fusion, and the pulse wave is classified as a C-type pulse wave. Alternatively, if w_t is not greater than the threshold, the features of the diabetic pre-wave and the diabetic wave are fused, the first peak is the main peak, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, the diabetic pre-wave is marked as the feature fusion, and the pulse wave is classified as a G type pulse wave. (4) Within the threshold range of the sampling point, the pulse wave has one peak, one single-point decreasing convex-concave point and another peak from left to right. The first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, the diabetic pre-wave is marked as the single-point decreasing convex-concave point, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, and the pulse wave is classified as a type D pulse wave. (5) Within the threshold range of the sampling point, the pulse wave has one peak and two single-point decreasing convex and concave points from left to right; then the first peak is the main peak, the first single-point decreasing convex and concave point is the diphtheria pre-wave and is marked as a single-point decreasing convex and concave point, the second single-point decreasing convex and concave point is the diphtheria wave and is marked as a single-point decreasing convex and concave point, and the pulse wave is classified as a type E pulse wave. (6) Within the threshold range of the sampling point, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right; then the first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, and it is marked as the single-point decreasing convex-concave point. The features of the diabetic wave and the diabetic pre-wave are fused, and the diabetic wave is marked as feature fusion. The pulse wave is classified as a type F pulse wave. (7) Within the threshold range of the sampling point, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right; then the first peak is the main peak, the single-point decreasing convex-concave point is the diabetic wave, and it is marked as the single-point decreasing convex-concave point. The diabetic pre-wave and diabetic wave features are fused, the diabetic pre-wave is marked as feature fusion, and the pulse wave is classified as H type pulse wave. (8) If there is a peak in the pulse wave from left to right within the threshold range of the sampling point, then the peak is the main peak. Both the diabetic pre-wave and the diabetic wave are marked as feature fusion, and the pulse wave is classified as a type I pulse wave.

[0008] Preferably, based on the classification results of the classification module, the temporal characteristic parameters of the pulse wave are calculated using the following method: For Class A pulse waves, to the right of the main peak, calculate the peaks and troughs of the diabetic pre-wave and the diabetic wave, where the first set of peaks and troughs represents the peaks and troughs of the diabetic pre-wave, and the second set of peaks and troughs represents the peaks and troughs of the diabetic wave. For type B pulse waves, calculate the peak and trough points to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. Calculate the single-point decreasing convex and concave points, where the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic wave. For a type C pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. The diabetic wave and diabetic pre-wave are then merged. For a type D pulse wave, calculate the peak and trough points to the right of the main peak to obtain the peak and trough of the diabetic wave; between the diabetic wave and the main peak, calculate the peak and trough points of the diabetic pre-wave, and the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic pre-wave. For a Class E pulse wave, calculate the peak and trough of the diabetic pre-wave and the diabetic wave to the right of the main peak; the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points from the main peak are the trough and peak of the diabetic pre-wave, and the maximum absolute value of the second set of concave points and the maximum absolute value of the convex points are the trough and peak of the diabetic wave. For type F pulse waves, the peak and trough points of the diabetic pre-wave are calculated to the right of the main peak. The trough and peak of the diabetic pre-wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic wave and diabetic pre-wave are then merged. For a type G pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic wave, and the diabetic pre-wave and diabetic wave are merged. For type H pulse waves, the peak and trough points of the diabetic wave are calculated to the right of the main peak. The trough and peak of the diabetic wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic pre-wave and diabetic wave are then merged. For Class I pulse waves, calculate the position of the main peak, and merge the diabetic wave and diabetic pre-wave with the main peak; Among them, the method for calculating the peak and trough points in the type A pulse wave is the extreme value method; And / or, in the type B pulse wave, the peak and trough points are calculated using the extreme value method, and the single-point decreasing convex and concave points are calculated using the curvature method; And / or, in the C-type pulse wave, the peak and trough points are calculated using the extreme value method; And / or, in the type D pulse wave, the peak and trough points are calculated using the extreme value method, and the single-point decreasing convex and concave points are calculated using the curvature method; And / or, in the E-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the F-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the G-type pulse wave, the peak and trough points are calculated using the extreme value method; And / or, in the H-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the Type I pulse wave, the method for calculating the peak and trough points is the extreme value method.

[0009] The preferred method for calculating the time-domain characteristic parameters of type A to type I pulse waves is as follows: The calculation method of time domain characteristic parameters of Class A pulse wave: (1) Calculate the coordinates of all peak points and valley points of the pulse wave by means of the pulse wave extreme value method; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the position of the valley point to the left of each peak point except the main wave peak, form a peak-valley pair, and calculate the height difference between the peak and valley; (4) Determine the position of the diabetic pre-wave and diabetic wave: form a pair of peak-valley pairs by forming adjacent valleys and peaks, and define the two points with the largest distance between the peak and valley pairs as the diabetic pre-wave and diabetic wave from left to right according to their coordinate positions; The calculation method of time-domain characteristic parameters of type B pulse wave is as follows: (1) Calculate all peak points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing points by the curvature method; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is obtained; the horizontal coordinate position corresponding to the peak point is x c Let the two points corresponding to the decrease of 1 / 3 in the peak value of the main wave be a and b, and let the x-coordinate of point b be x. b Within the sampling point threshold range (x) c， x b), and define the point with the largest difference between the peak and the valley as the coordinate of the diabetic pre-wave; (5) Determine the diabetic wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a convex and concave single-point decreasing convex and concave point pair, and calculate the height difference between the convex and concave single-point decreasing convex and concave point pairs, e is the peak of the diabetic pre-wave, and the horizontal coordinate corresponding to point e is x. e Within the threshold range of sampling points (x) e， (0.7N), the point with the largest difference is defined as the coordinate of the diphtheria wave; N is the total number of sampling points; The calculation method of time-domain characteristic parameters of C-type pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. The sampling point threshold range is 1-0.7N. The point with the largest difference between the peak and valley is defined as the coordinate of the diabetic anterior wave; N is the total number of sampling points. The calculation method of time-domain characteristic parameters of type D pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is obtained. e is the peak of the diabetic wave, and the horizontal coordinate of point e is x. e Within the threshold range of sampling points (x) e， 0.7N), the point with the largest difference between the peak and valley is defined as the coordinate of the diabetic wave; N is the total number of sampling points; (5) Determine the diabetic wave: the position of the concave single point decreasing convex and concave point to the left of each convex single point decreasing convex and concave point constitutes a pair of convex and concave single points decreasing convex and concave points, and the height difference between the pairs of convex and concave single points decreasing convex and concave points is calculated; within the range of the horizontal coordinates of the diabetic wave valley and the main wave peak, the point with the largest difference is defined as the coordinate of the diabetic wave; The calculation method of time domain characteristic parameters of E-type pulse wave: (1) Calculate the coordinates of all peak points and valley points of pulse wave through the extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point within the range to the right of the main wave peak, form a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; (5) Determine the position of diabetic pre-wave and diabetic wave: define the two points with the largest height difference of convex and concave single-point decreasing convex and concave points from left to right according to their coordinate positions as diabetic pre-wave and diabetic wave; The calculation method of time-domain characteristic parameters of type F pulse wave is as follows: (1) Calculate the coordinates of all peak points and valley points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of the pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic pre-wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of the diabetic wave and the main wave peak, the point with the largest difference is defined as the coordinate of the diabetic pre-wave; The calculation method of time-domain characteristic parameters of type G pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinate of the largest peak point as the position of the main wave peak; (3) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. e is the peak of the diabetic wave, and the horizontal coordinate of point e is x. e Within the threshold range of sampling points (x) e， 0.7N), the point with the largest difference between the peak and the valley is defined as the coordinate of the diphtheria wave; N is the total number of sampling points; The calculation method of time-domain characteristic parameters of type H pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of diabetic wave and main wave peak, the point with the largest difference is defined as the coordinate of diabetic wave; The calculation method of time-domain characteristic parameters of type I pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak.

[0010] The present invention also provides a pulse wave time-domain feature recognition system, comprising: The input module is configured to input a pulse wave; The prediction module integrates the aforementioned pulse wave time-domain feature classification system. The classification module is configured to identify the pulse wave through a machine learning model within the threshold range of the sampling points, classifying the dicrotic anterior wave into three features: extreme points, single-point decreasing convex and concave points, and feature fusion, and classifying the dicrotic wave into three features: extreme points, single-point decreasing convex and concave points, and feature fusion. Based on the classification results of the dicrotic anterior wave and the dicrotic wave, the pulse wave is classified into Class A to Class I pulse waves. The output module is configured to output the classification result of the prediction module.

[0011] Preferably, the pulse wave is a single-cycle optimal pulse wave; the three characteristics of the dicrotic pre-wave and the three characteristics of the dicrotic wave are calculated using the single-cycle optimal pulse wave.

[0012] Preferably, the machine learning model is selected from a convolutional neural network feature recognition model.

[0013] Preferably, the convolutional neural network feature recognition model includes a convolutional neural network feature recognition model for the dicrotic anterior wave and a convolutional neural network feature recognition model for the dicrotic wave; the two models are used to identify the types of the dicrotic anterior wave and the dicrotic wave of the pulse wave, respectively, and determine that the pulse wave is one of the nine types of waveforms from type A to type I pulse waves.

[0014] Preferably, the prediction module is further configured to calculate the temporal characteristic parameters of the pulse wave based on the classification result of the pulse wave.

[0015] The present invention also provides a computer-readable storage medium having stored thereon a computer program for implementing the above-described classification system, or a computer program for implementing the above-described identification system.

[0016] A pulse wave mainly consists of the main wave, the dicrotic pre-wave, and the dicrotic wave. The dicrotic pre-wave is classified into three features: extreme points, single-point decreasing convex / concave points, and feature fusion. The dicrotic wave is also classified into three features: extreme points, single-point decreasing convex / concave points, and feature fusion. Based on the calculation results, the pulse wave is classified into nine waveform types, from Class A to Class I. This invention provides a pulse wave time-domain feature recognition system based on morphological features. This system utilizes a machine learning model to identify the three features of the dicrotic pre-wave and the dicrotic wave by inputting the best single-cycle pulse wave, achieving classification of the nine waveform types from Class A to Class I. Based on the classification characteristics of the dicrotic pre-wave and the dicrotic wave, the system uses the extreme value method or curvature method to obtain the time-domain feature parameters of Class A to Class I pulse waves, improving the accuracy of pulse wave time-domain feature extraction and classification precision. This allows doctors to more accurately diagnose diseases based on the morphological characteristics of the pulse wave.

[0017] This invention classifies pulse waves into nine categories based on Tibetan medicine diagnostic principles. After obtaining the classification results of a patient's pulse wave, this information can assist Tibetan medicine in diagnosing diseases (related to the characteristics of the main wave, pre-diplastic wave, and diplastic wave). For example, in the diagnosis of hypertension, if a patient's pulse wave is classified as E or F, the patient's risk of developing the disease can be determined to be higher; conversely, if the pulse wave is classified as A or G, the patient's risk of developing the disease can be determined to be lower.

[0018] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.

[0019] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description

[0020] Figure 1 This is a standard single-cycle pulse wave pattern, where the two points corresponding to the 1 / 3 decrease in the peak value of the main wave are points a and b; c is the main wave peak, d is the trough of the dicrotic wave, e is the peak of the dicrotic wave, f is the trough of the dicrotic wave, g is the peak of the dicrotic wave, h1 is the height of the main wave peak, h3 is the height of the peak of the dicrotic wave, h4 is the height of the trough of the dicrotic wave, h5 is the height difference between the peak and trough of the dicrotic wave, t1 is the time distance from the pulse wave initiation point to the main wave peak, t4 is the time distance from the pulse wave initiation point to the trough of the dicrotic wave, t5 is the time distance from the trough of the dicrotic wave to the end point of the pulse wave, and t is the pulse wave time period.

[0021] Figure 2 The characteristics and waveform morphology of pulse waves from Class A to Class E are described.

[0022] Figure 3 The characteristics and waveform morphology of pulse waves from type F to type I are described.

[0023] Figure 4 This is the confusion matrix of the convolutional neural network feature recognition model for the diabetic pre-wave.

[0024] Figure 5 The confusion matrix is ​​the convolutional neural network feature recognition model for diphtheria waves. Detailed Implementation

[0025] It should be noted that the algorithms for data acquisition, transmission, storage, and processing steps not specifically described in the embodiments can all be implemented using publicly available information in the prior art.

[0026] Example 1: Pulse Wave Time-Domain Feature Classification System The pulse wave time-domain feature classification system includes a classification module and a calculation module. The classification module is configured to classify pulse waves according to the following morphological features: Type A pulse waves include the following morphological characteristics: the morphological characteristics of both the dicrotic pre-wave and the dicrotic wave are extreme points; Type B pulse waves include the following morphological characteristics: the morphological characteristics of the dicrotic pre-wave are extreme points, and the morphological characteristics of the dicrotic wave are single-point decreasing convex and concave points. Type C pulse waves include the following morphological characteristics: the morphological characteristic of the dicrotic pre-wave is an extreme point, and the morphological characteristic of the dicrotic wave is feature fusion; Type D pulse waves include the following morphological characteristics: the morphological characteristics of the dicrotic pre-wave are single-point decreasing convex and concave points, and the morphological characteristics of the dicrotic wave are extreme points. Type E pulse waves include the following morphological characteristics: both the dicrotic pre-wave and the dicrotic wave are characterized by single-point decreasing convex and concave points. Type F pulse waves include the following morphological characteristics: the morphological characteristics of the dicrotic pre-wave are single-point decreasing convex and concave points, and the morphological characteristics of the dicrotic wave are feature fusion. Type G pulse waves include the following morphological characteristics: the morphological characteristics of the dicrotic pre-wave are feature fusion, and the morphological characteristics of the dicrotic wave are extreme points; Type H pulse waves include the following morphological characteristics: the morphological characteristics of the dicrotic pre-wave are feature fusion, and the morphological characteristics of the dicrotic wave are single-point decreasing convex and concave points. Type I pulse waves include the following morphological features: the morphological features of the dicrotic pre-wave and the dicrotic wave are both feature fusions.

[0027] The methods for determining extreme points, single-point decreasing convex and concave points, or feature fusion in the classification module are as follows: (1) The threshold range of the sampling points is 1-0.7N, where N is the number of sampling points. Within the threshold range of the sampling points, there are three peaks in the pulse wave. The first peak is the main peak, the second peak is the peak of the diabetic pre-wave, and the valley to its left is the valley of the diabetic pre-wave. The third peak is the peak of the diabetic wave, and the valley to its left is the valley of the diabetic wave. The diabetic pre-wave and diabetic wave are marked as extreme points respectively. (2) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has two peaks from left to right and a single-point decreasing convex-concave point. The first peak is the main wave peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point. The single-point decreasing convex-concave point is the diabetic wave. The diabetic wave is marked as the single-point decreasing convex-concave point. (3) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has two peaks from left to right, and there are no single-point decreasing convex or concave points; such as Figure 1 As shown, the two points corresponding to the 1 / 3 decrease in the peak value of the main wave are defined as points a and b. The difference between the sampling point coordinates corresponding to point b and the sampling point coordinates corresponding to point a is the pulse width w. The ratio of the pulse width to the total number of sampling points N is w_t. If w_t is greater than the threshold, the features of the diabetic wave and the diabetic pre-wave are fused. The first peak is the main peak, and the second peak is the peak of the diabetic pre-wave. The diabetic pre-wave is marked as the extreme point, and the diabetic wave is marked as the feature fusion. Alternatively, if w_t is not greater than the threshold, the features of the diabetic pre-wave and the diabetic wave are fused, the first peak is the main peak, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, and the diabetic pre-wave is marked as the feature fusion. (4) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has one peak, one single-point decreasing convex-concave point and another peak from left to right. The first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, and the diabetic pre-wave is marked as the single-point decreasing convex-concave point. The second peak is the diabetic wave peak, and the diabetic wave is marked as the extreme point. (5) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has one peak and two single-point decreasing convex and concave points from left to right. The first peak is the main peak, the first single-point decreasing convex and concave point is the diphtheria pre-wave and is marked as a single-point decreasing convex and concave point, and the second single-point decreasing convex and concave point is the diphtheria wave and is marked as a single-point decreasing convex and concave point. (6) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right. The first peak is the main peak, and the single-point decreasing convex-concave point is the diabetic pre-wave. It is marked as the single-point decreasing convex-concave point. The diabetic wave and the diabetic pre-wave feature are fused, and the diabetic wave is marked as feature fusion. (7) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right. The first peak is the main peak, and the single-point decreasing convex-concave point is the diabetic wave. It is marked as the single-point decreasing convex-concave point. The diabetic pre-wave and diabetic wave features are fused, and the diabetic pre-wave is marked as feature fusion. (8) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, if there is a peak in the pulse wave from left to right, then the peak is the main peak. Both the diabetic pre-wave and the diabetic wave are marked as feature fusion.

[0028] The calculation module is configured to calculate the temporal characteristic parameters of the pulse wave based on the classification results from the classification module, using the following method: For Class A pulse waves, to the right of the main peak, calculate the peaks and troughs of the diabetic pre-wave and the diabetic wave, where the first set of peaks and troughs represents the peaks and troughs of the diabetic pre-wave, and the second set of peaks and troughs represents the peaks and troughs of the diabetic wave. For type B pulse waves, calculate the peak and trough points to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. Calculate the single-point decreasing convex and concave points, where the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic wave. For a type C pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. The diabetic wave and diabetic pre-wave are then merged. For a type D pulse wave, calculate the peak and trough points to the right of the main peak to obtain the peak and trough of the diabetic wave; between the diabetic wave and the main peak, calculate the peak and trough points of the diabetic pre-wave, and the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic pre-wave. For a Class E pulse wave, calculate the peak and trough of the diabetic pre-wave and the diabetic wave to the right of the main peak; the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points from the main peak are the trough and peak of the diabetic pre-wave, and the maximum absolute value of the second set of concave points and the maximum absolute value of the convex points are the trough and peak of the diabetic wave. For type F pulse waves, the peak and trough points of the diabetic pre-wave are calculated to the right of the main peak. The trough and peak of the diabetic pre-wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic wave and diabetic pre-wave are then merged. For a type G pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic wave, and the diabetic pre-wave and diabetic wave are merged. For type H pulse waves, the peak and trough points of the diabetic wave are calculated to the right of the main peak. The trough and peak of the diabetic wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic pre-wave and diabetic wave are then merged. For Class I pulse waves, calculate the position of the main peak, and merge the diabetic wave and diabetic pre-wave with the main peak; Specifically, the calculation method for the time-domain characteristic parameters of the pulse wave from Class A to Class I is as follows: The calculation method for the time-domain characteristic parameters of the pulse wave is as follows: (1) Calculate the coordinates of all peak points and trough points of the pulse wave using the pulse wave extreme value method; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the position of the trough point to the left of each peak point except the main wave peak, form a peak-trough pair, and calculate the height difference between the peak and the trough; (4) Determine the position of the diabetic pre-wave and diabetic wave: form a peak-trough pair with adjacent troughs and peaks, and define the two points with the largest distance between the peak and trough pairs as the diabetic pre-wave and diabetic wave from left to right according to their coordinate positions. The calculation method of time-domain characteristic parameters of type B pulse wave is as follows: (1) Calculate all peak points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing points by the curvature method; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is obtained; the horizontal coordinate position corresponding to the peak point is x c, The x-coordinate of point b is x. b, Within the sampling point threshold range (x) c， x b ), and define the point with the largest difference between the peak and the valley as the coordinate of the diabetic anterior wave; (5) Determine the diabetic wave: the position of the concave single point decreasing to the left of each convex single point decreasing ... e Within the threshold range of sampling points (x) e， (0.7N), the point with the largest difference is defined as the coordinate of the diphtheria wave; The calculation method of time-domain characteristic parameters of C-type pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. The sampling point threshold range is 1-0.7N. The point with the largest difference between the peak and valley is defined as the coordinate of the diabetic anterior wave. The calculation method of time-domain characteristic parameters of type D pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. Within the threshold range of the sampling point (x e， 0.7N), the point with the largest difference between the peak and valley is defined as the coordinate of the diabetic wave; (5) Determine the diabetic anterior wave: the position of the concave single point decreasing to the left of each convex single point decreasing ... The calculation method of time domain characteristic parameters of E-type pulse wave: (1) Calculate the coordinates of all peak points and valley points of pulse wave through the extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point within the range to the right of the main wave peak, form a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; (5) Determine the position of diabetic pre-wave and diabetic wave: define the two points with the largest height difference of convex and concave single-point decreasing convex and concave points from left to right according to their coordinate positions as diabetic pre-wave and diabetic wave; The calculation method of time-domain characteristic parameters of type F pulse wave is as follows: (1) Calculate the coordinates of all peak points and valley points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of the pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic pre-wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of the diabetic wave and the main wave peak, the point with the largest difference is defined as the coordinate of the diabetic pre-wave; The calculation method of time-domain characteristic parameters of G-type pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. Within the threshold range of the sampling point (x e，0.7N), the point with the largest difference between the peak and trough is defined as the coordinate of the diphtheria wave; The calculation method of time-domain characteristic parameters of type H pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of diabetic wave and main wave peak, the point with the largest difference is defined as the coordinate of diabetic wave; The calculation method of time-domain characteristic parameters of type I pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak.

[0029] The method for calculating the extreme points of the pulse wave is as follows: Assuming the data length of a single-cycle pulse wave y(n) is N, the method for determining if the i-th point is a peak point is as follows: By iterating through all sampling points on y(n), the location of the pulse wave maximum can be obtained using the above formula; By iterating through all sampling points on y(n), the location of the pulse wave minimum can be obtained using the above formula; The method for calculating the convex and concave points of a single-point decrease in the pulse wave is as follows: Based on the definition of curvature, obtain the curvature transformation expression: in the formula This represents the curvature sequence values ​​of y(n). Let y be the first derivative. Let y be the second derivative. In order to remove the interference terms in the curvature, the original curvature is subjected to a 5-point smoothing filter. The number of filter points is determined by the sampling frequency of the original data. By iterating through all sampling points on k(n), the above formula can be used to obtain the position of the convex and concave points of the pulse wave single-point decrease. By iterating through all sampling points on k(n), the position of the concave point of the pulse wave with single-point decrease can be obtained using the above formula.

[0030] Example 2: Pulse Wave Time Domain Feature Recognition System The input module is configured to input the optimal single-cycle pulse wave. The prediction module integrates the pulse wave time-domain feature classification system described in Example 1. It is configured to identify pulse waves through a machine learning model, classify the dicrotic anterior wave into three features: extreme points, single-point decreasing convex and concave points, and feature fusion, classify the dicrotic wave into three features: extreme points, single-point decreasing convex and concave points, and feature fusion, and classify the pulse wave into nine waveforms from Class A to Class I based on the calculation results. The output module is configured to output the classification result of the prediction module.

[0031] The calculation method for the optimal single-cycle pulse wave in the input module is as follows: Based on the morphology of the pulse wave, a motion interference recognition algorithm is proposed. Here, the expression of the pulse wave is set as x(n), and the total data length of the pulse wave is N. The steps are as follows: (1) Baseline drift removal: Baseline drift will cause the pulse wave peak to fluctuate. In order to eliminate the influence of baseline drift, a second-order mean filter is used to remove the baseline drift. Assuming that the average window length of the first stage is L1 and the average window length of the second stage is L2, the envelope waveform after removing the baseline at point j can be obtained using formulas (6) and (7). In the formula, max{} is the function for finding the maximum value. N 1 represents the left-end index of the window given an average window length of L1; N 2 is the index of the right end of the window given an average window length of L1; N 3 represents the index of the left end of the window, given an average window length of L2. N 4 represents the left-end index of the window given an average window length of L2. x 1() represents the original signal. x 2() represents the processed signal. i To calculate the range index of sample points in the summation formula, j This is the index of the current sample point.

[0032] (2) Under certain circumstances, the frequency bands of physiological signals and noise overlap, but the noise is relatively weak and the noise signal is random, while the pulse wave signal is repetitive. For a segment of pulse wave data x, T is truncated. max The length of the pulse wave waveform, with a set step size T. min Find the difference d(i) between the maximum and minimum values ​​of the pulse wave for each data segment, and then find the median of all d values. The calculation formula is as follows: In the formula, M represents the maximum number of pulse wave segments to be extracted, i represents the i-th data segment, and med is the median function. Since the pulse wave amplitude fluctuates within a specified range (±0.5 times), assuming that for a pulse wave x(n), the number of pulse waves in a single cycle is k, and the x-coordinate of the starting point of a single cycle pulse wave is... ,in Let represent the termination point of the k-th pulse wave. For a single-cycle pulse wave, the following equation should be satisfied: d(m) represents the difference between the maximum and minimum values ​​of the m-th pulse wave, and x(a) represents the periodic sequence of the a-th pulse wave. The sequence number of the starting point of the pulse wave that satisfies the above conditions is stored in N. num middle; (3) The position of the peak of a single pulse wave is within a range, and its rise time is shorter than its fall time. For the first pulse wave... The peak point of each pulse wave is located at... The location of the pulse wave starting point is Then the position of the pulse wave starting point should satisfy: Save the sequence number of the pulse wave starting point that meets the above conditions in N. num middle; (4) For a single-cycle pulse wave, the first-order difference of the waveform within the range from the pulse wave initiation point to the pulse wave peak point is positive, and its maximum value is First-order difference express; In the formula, x(m) represents the sequence value of the m-th pulse wave, and m represents the x-coordinate from the starting point of the pulse wave. x-axis of the pulse wave peak point Position sequence.

[0033] Find the maximum absolute value of the first-order difference of the pulse wave within a certain range near the termination point of the pulse wave. First-order difference express; In the formula, abs() means taking the absolute value. Indicates from point k to Multiple corresponding pulse wave sequence values.

[0034] Find the maximum value of the first-order difference. The first-order difference value of a single-cycle pulse wave should satisfy the following condition: Save the sequence number of the pulse wave starting point that meets the above conditions in N. num middle; (5) As shown in the waveform diagram of the pulse wave, w is the data width at 1 / 3 of the main wave of the pulse wave. For a single cycle of the pulse wave, assume that w is at two points on the pulse wave, the point closest to the starting point. Within the range from the pulse wave initiation point to point a, there is no peak point, that is: In the formula, This represents the x-coordinate position of point a corresponding to the i-th pulse wave. The position of point a is as follows: Figure 1 As shown.

[0035] (6) In the above pulse wave recognition process, there may still be jitter misjudgment. For a pulse wave of a data segment, according to the pulse wave periodicity characteristics, it is necessary to ensure that three consecutive data segments are pulse waves, then the data segment can be determined to be a pulse wave. (7) Optimal pulse wave extraction method: The average number of sampling points of n pulse waves in a pressure range is used. The length of the pulse wave is unified as the average number of sampling points using third-order spline interpolation. The coherent averaging algorithm is used to average the n pulse waves point by point to obtain the optimal pulse wave for a single cycle.

[0036] In the prediction module, the morphological characteristics of the diabetic anterior wave and diabetic wave, and the annotation methods for the three features are as follows: Based on the temporal characteristics of the diabetic pre-wave and diabetic wave, the diabetic pre-wave and diabetic wave are respectively labeled as three features: extreme value, single-point decreasing convex and concave point, and feature fusion. The characteristic morphology of the main wave can be divided into one state: extreme point; the characteristic morphology of the dicrotic pre-wave and dicrotic wave can be divided into three states: extreme point, single-point decreasing convex and concave points, and characteristic fusion; by combining the three states of the dicrotic pre-wave and dicrotic wave, the pulse wave can be divided into nine categories. The morphological characteristics of the diabetic pre-wave and diabetic wave are mainly divided into three features: extreme points, single-point decreasing convex and concave points, and feature fusion. The diabetic pre-wave is an extreme point, characterized by the presence of peaks and troughs in the diabetic pre-wave; the diabetic pre-wave is a single-point decreasing convex-concave point, characterized by a monotonically decreasing diabetic pre-wave curve, where concave points are diabetic pre-wave troughs and convex points are diabetic pre-wave peaks; the diabetic pre-wave is a feature fusion, characterized by the fusion of diabetic pre-wave and diabetic wave features, or the fusion of diabetic pre-wave and main peak features. When the diphthalamic wave is an extreme point, it is characterized by the presence of peaks and troughs in the diphthalamic wave; when the diphthalamic wave is a single-point decreasing convex-concave point, it is characterized by the monotonically decreasing diphthalamic wave curve, where the concave point is the diphthalamic wave trough and the convex point is the diphthalamic wave peak; when the diphthalamic wave is a feature fusion, it is characterized by the fusion of the diphthalamic wave with the features of the diphthalamic wave preceding the wave, or the fusion of the diphthalamic wave with the features of the main peak. The method for labeling three features of diphtheria pre-wave and diphtheria wave is characterized by: (1) The threshold range of the sampling point is 1-0.7N, where N is the number of sampling points. Within the threshold range of the sampling point, there are three peaks in the pulse wave. The first peak is the main peak, the second peak is the diphtheria pre-wave peak, and the valley to its left is the diphtheria pre-wave valley. The third peak is the diphtheria wave peak, and the valley to its left is the diphtheria wave valley. The diphtheria pre-wave and diphtheria wave are labeled as extreme points respectively. (2) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has two peaks from left to right and a single-point decreasing convex-concave point. The first peak is the main wave peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point. The single-point decreasing convex-concave point is the diabetic wave. The diabetic wave is marked as the single-point decreasing convex-concave point. (3) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, there are two peaks from left to right in the pulse wave, and there are no single point of decreasing convex and concave points. The two points corresponding to the decrease of 1 / 3 of the peak value of the main wave are a and b. The difference between the sampling point coordinate value corresponding to point b and the sampling point coordinate value corresponding to point a is the pulse width w. The ratio of the pulse width to the total number of sampling points N is w_t. If w_t is greater than the threshold, the features of the diabetic wave and the diabetic pre-wave are fused. The first peak is the main peak, and the second peak is the peak of the diabetic pre-wave. The diabetic pre-wave is marked as the extreme point, and the diabetic wave is marked as the feature fusion. Alternatively, if w_t is not greater than the threshold, the features of the diabetic pre-wave and the diabetic wave are fused, the first peak is the main peak, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, and the diabetic pre-wave is marked as the feature fusion. (4) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has one peak, one single-point decreasing convex-concave point and another peak from left to right. The first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, and the diabetic pre-wave is marked as the single-point decreasing convex-concave point. The second peak is the diabetic wave peak, and the diabetic wave is marked as the extreme point. (5) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has one peak and two single-point decreasing convex and concave points from left to right. The first peak is the main peak, the first single-point decreasing convex and concave point is the diphtheria pre-wave and is marked as a single-point decreasing convex and concave point, and the second single-point decreasing convex and concave point is the diphtheria wave and is marked as a single-point decreasing convex and concave point. (6) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right. The first peak is the main peak, and the single-point decreasing convex-concave point is the diabetic pre-wave. It is marked as the single-point decreasing convex-concave point. The diabetic wave and the diabetic pre-wave feature are fused, and the diabetic wave is marked as feature fusion. (7) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right. The first peak is the main peak, and the single-point decreasing convex-concave point is the diabetic wave. It is marked as the single-point decreasing convex-concave point. The diabetic pre-wave and diabetic wave features are fused, and the diabetic pre-wave is marked as feature fusion. (8) The threshold range of the sampling points is 1-0.7N, where N is the total number of sampling points. Within the threshold range of the sampling points, if there is a peak in the pulse wave from left to right, then the peak is the main peak. Both the diabetic pre-wave and the diabetic wave are marked as feature fusion.

[0037] In the prediction module, the network structure and training method of the machine learning model are as follows: Two feature recognition models based on convolutional neural networks were constructed using three features of the diphtheria wave and the pre-diphtheria wave as the training and test sets, respectively, to identify the three features of the diphtheria wave and the pre-diphtheria wave. One-dimensional convolutional neural networks (1D-CNNs) consist of an input layer, a convolutional layer (CONV), a fully connected layer (FC), and an output layer. The main factors affecting the recognition accuracy and training time of 1D-CNN models are the sample size, model structure, and hyperparameters. Since the number of samples used in the experiment is limited, the recognition of single-cycle waveforms is achieved by changing the model structure and hyperparameters. Using three characteristic waveforms of the diabetic anterior wave as input, during the training process, we tried to change the learning rate, the number of Conv layers, and the length of the input data. By comparing the performance of the validation set, we found the hyperparameter combination and model structure that best performed the task. Based on this, we shortened the training time and obtained the best convolutional neural network feature recognition model for the diabetic anterior wave. Using three characteristic waveforms of the diphtheria wave as input, during training, we tried changing the learning rate, the number of Conv layers, and the length of the input data. By comparing the performance on the validation set, we found the hyperparameter combination and model structure that best performed the task. Based on this, we shortened the training time and obtained the best convolutional neural network feature recognition model for the diphtheria wave.

[0038] Let the convolutional neural network feature recognition model for the dicrotic pre-wave be model A, and the convolutional neural network feature recognition model for the dicrotic wave be model B. Inputting a single-cycle pulse wave signal, the types of the dicrotic pre-wave and dicrotic wave are identified using models A and B. Based on these types, the pulse wave is determined to be one of nine waveform types from type A to type I (e.g.,...). Figure 2 , Figure 3 (As shown).

[0039] The output module outputs the classification result of the prediction module, specifically obtaining the classification result and the temporal characteristic parameters of the pulse wave in Example 1.

[0040] The technical solution of the present invention will be further explained through experiments below.

[0041] Experiment Example 1: Evaluation Results of Pulse Wave Feature Recognition Model The convolutional neural network feature recognition models for the diphtheria anterior wave and the diphtheria wave used in this experimental example were calculated according to Example 1.

[0042] I. Experimental Methods A total of 150 volunteers participated in the data collection process. The volunteers were team members from the laboratory, including 86 male and 64 female participants. Throughout the experiment, each participant fully understood the research content and voluntarily participated in data collection. The specific inclusion criteria for the participant sample are shown in Table 1.

[0043] ① Cardiovascular health, with no cardiovascular-related diseases; ② In good physical condition, with no acute illness; ③ I have not recently taken any medications that affect heart rate or cause vasoconstriction.

[0044] A total of 1,623 single-cycle waveforms were collected. Among them, the number of diphtheria wavefront extreme points, single-point decreasing convex and concave points, and feature fusions were 537, 559, and 527, respectively; the number of diphtheria waves extreme points, single-point decreasing convex and concave points, and feature fusions were 641, 532, and 450, respectively.

[0045] Table 1. Physiological parameters of volunteers II. Experimental Results 1. Performance evaluation of the convolutional neural network feature recognition model for diphtheria pre-wave As shown in Table 3, after training with the 1D-CNN model, the overall accuracy on the test set was 95.06%, and the overall classification accuracy of the model reached over 94.91%, indicating that the model has a highly stable classification ability for the three features of diphtheria pre-wave and high classification accuracy; the macro-F1 accuracy on the validation set and test set was 93.04% and 94.91%, respectively (as shown in Tables 2 and 3). Figure 4 As shown in the figure, this indicates that the model's overall classification performance is relatively balanced across all categories.

[0046] Table 2. Validation set results of the convolutional neural network feature recognition model for diabetic pre-wave. Table 3. Test set results of the convolutional neural network feature recognition model for diabetic pre-wave. 2. Performance evaluation of the convolutional neural network feature recognition model for diphtheria waves As shown in Table 5, after training with the 1D-CNN model, the overall accuracy on the test set is 93.70%, and the overall classification accuracy of the model reaches over 93.60%, indicating that the model has a highly stable classification ability for the three features of diphtheria waves and high classification accuracy; the macro-F1 accuracy on the validation set and test set is 97.23% and 93.70% respectively (as shown in Tables 4 and 5). Figure 5 As shown in the figure, this indicates that the model's overall classification performance is relatively balanced across all categories.

[0047] Table 4. Validation set results of the convolutional neural network feature recognition model for diphtheria waves. Table 5. Test set results of the convolutional neural network feature recognition model for diphtheria waves. Experiment Example 2: Evaluation Results of Pulse Wave Time-Domain Feature Recognition System The pulse wave time-domain feature recognition system used in this experimental example was prepared according to Example 2. I. Experimental Methods Correction determination coefficient (Adjusted R-Square): This parameter represents the ratio of the regression sum of squares to the residual sum of squares, a measure of the goodness of fit of the regression. The closer this statistic is to 1, the better the model fit, and it is an important indicator of accuracy in regression problems; Mean Absolute Error (MAO) ): This parameter represents the average of the sum of the absolute values ​​of the differences between the predicted and actual values. This statistic better reflects the actual situation of the prediction error; Root Mean Square Error (RMSE) ): This parameter is the square root of the mean square error; , The smaller the value, the higher the prediction accuracy of the model. The calculation method is as follows: in, Indicates the predicted value. Indicates the actual value. denoted as the average of the actual values, and n represents the sample size; 100 diphtheria pre-wave and diphtheria wave are randomly selected as extreme points, convex and concave points with single-point decreasing, and pulse waves with feature fusion for evaluating the identification results of diphtheria wave and diphtheria pre-wave coordinate points; the accuracy of the algorithm is evaluated by the algorithm calculation results of the three states of diphtheria pre-wave and diphtheria wave.

[0048] II. Experimental Results As can be seen from the results in Table 6, after using the classification and recognition algorithm of this invention, the lowest corrected regression coefficients for the three features of the pulse wave time-domain diphtheria wave front wave are 0.97, the lowest corrected coefficient for the diphtheria wave is 0.96, and the statistical parameters such as MAE and RMSE corresponding to each feature point dataset are less than 2, indicating that the pulse wave time-domain feature recognition system has achieved high feature point recognition accuracy.

[0049] Table 6. Results of Diphasic Wave and Diphasic Pre-Wave Coordinate Point Recognition The experimental results above show that the feature recognition models for diphtheria pre-wave and diphtheria wave constructed in this invention have highly stable classification ability and high classification accuracy for the three features, and the classification performance for each category is relatively balanced. At the same time, the pulse wave time-domain feature recognition system of this invention has high accuracy and can guide the diagnosis of various diseases in Tibetan medicine.

Claims

1. A pulse wave time-domain feature classification system, characterized in that, include: The sampling module is configured to collect sampling point data, with a threshold range of 1-0.7N for the sampling points, where N is the total number of sampling points. The classification module is configured to identify the morphological features of the dicrotic pre-wave and dicrotic wave in the pulse wave within a threshold range of the sampling points, and classify the pulse wave into Class A to Class I pulse waves based on the morphological features. The morphological characteristics of pulse waves from Class A to Class I are as follows: (1) Within the threshold range of the sampling point, there are three peaks in the pulse wave. The first peak is the main peak, the second peak is the peak of the diabetic pre-wave, and the valley to its left is the valley of the diabetic pre-wave. The third peak is the peak of the diabetic wave, and the valley to its left is the valley of the diabetic wave. The diabetic pre-wave and diabetic wave are marked as extreme points, and the pulse wave is classified as a type A pulse wave. (2) Within the threshold range of the sampling point, the pulse wave has two peaks from left to right and a single point decreasing convex and concave point. The first peak is the main wave peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point. The single point decreasing convex and concave point is the diabetic wave. The diabetic wave is marked as the single point decreasing convex and concave point. The pulse wave is classified as a type B pulse wave. (3) Within the threshold range of the sampling points, there are two peaks in the pulse wave from left to right; the two points corresponding to the decrease of 1 / 3 of the peak value of the main wave are a and b. The difference between the sampling point coordinate value corresponding to point b and the sampling point coordinate value corresponding to point a is the pulse width w, and the ratio of the pulse width to the total number of sampling points N is w_t. If w_t is greater than the threshold, the features of the diabetic wave and the diabetic pre-wave are fused. The first peak is the main peak, and the second peak is the diabetic pre-wave peak. The diabetic pre-wave is marked as the extreme point, the diabetic wave is marked as the feature fusion, and the pulse wave is classified as a C-type pulse wave. Alternatively, if w_t is not greater than the threshold, the features of the diabetic pre-wave and the diabetic wave are fused, the first peak is the main peak, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, the diabetic pre-wave is marked as the feature fusion, and the pulse wave is classified as a G type pulse wave. (4) Within the threshold range of the sampling point, the pulse wave has one peak, one single-point decreasing convex-concave point and another peak from left to right. The first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, the diabetic pre-wave is marked as the single-point decreasing convex-concave point, the second peak is the diabetic wave peak, the diabetic wave is marked as the extreme point, and the pulse wave is classified as a type D pulse wave. (5) Within the threshold range of the sampling point, the pulse wave has one peak and two single-point decreasing convex and concave points from left to right; then the first peak is the main peak, the first single-point decreasing convex and concave point is the diphtheria pre-wave and is marked as a single-point decreasing convex and concave point, the second single-point decreasing convex and concave point is the diphtheria wave and is marked as a single-point decreasing convex and concave point, and the pulse wave is classified as a type E pulse wave. (6) Within the threshold range of the sampling point, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right; then the first peak is the main peak, the single-point decreasing convex-concave point is the diabetic pre-wave, and it is marked as the single-point decreasing convex-concave point. The features of the diabetic wave and the diabetic pre-wave are fused, and the diabetic wave is marked as feature fusion. The pulse wave is classified as a type F pulse wave. (7) Within the threshold range of the sampling point, the pulse wave has a peak and a single-point decreasing convex-concave point from left to right; then the first peak is the main peak, the single-point decreasing convex-concave point is the diabetic wave, and it is marked as the single-point decreasing convex-concave point. The diabetic pre-wave and diabetic wave features are fused, the diabetic pre-wave is marked as feature fusion, and the pulse wave is classified as H type pulse wave. (8) If there is a peak in the pulse wave from left to right within the threshold range of the sampling point, then the peak is the main peak. Both the diabetic pre-wave and the diabetic wave are marked as feature fusion, and the pulse wave is classified as a type I pulse wave.

2. The pulse wave time-domain feature classification system according to claim 1, characterized in that, It also includes a calculation module, configured to calculate the temporal characteristic parameters of the pulse wave based on the classification results of the classification module, using the following method: For Class A pulse waves, to the right of the main peak, calculate the peaks and troughs of the diabetic pre-wave and the diabetic wave, where the first set of peaks and troughs represents the peaks and troughs of the diabetic pre-wave, and the second set of peaks and troughs represents the peaks and troughs of the diabetic wave. For type B pulse waves, calculate the peak and trough points to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. Calculate the single-point decreasing convex and concave points, where the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic wave. For a type C pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic pre-wave. The diabetic wave and diabetic pre-wave are then merged. For a type D pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic wave; between the diabetic wave and the main peak, the peak and trough points of the diabetic pre-wave are calculated, and the maximum absolute value of the concave point and the maximum absolute value of the convex point are the trough and peak of the diabetic pre-wave. For a Class E pulse wave, calculate the peak and trough of the diabetic pre-wave and the diabetic wave to the right of the main peak; the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points from the main peak are the trough and peak of the diabetic pre-wave, and the maximum absolute value of the second set of concave points and the maximum absolute value of the convex points are the trough and peak of the diabetic wave. For type F pulse waves, the peak and trough points of the diabetic pre-wave are calculated to the right of the main peak. The trough and peak of the diabetic pre-wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic wave and diabetic pre-wave are then merged. For a type G pulse wave, the peak and trough points are calculated to the right of the main peak to obtain the peak and trough of the diabetic wave, and the diabetic pre-wave and diabetic wave are merged. For type H pulse waves, the peak and trough points of the diabetic wave are calculated to the right of the main peak. The trough and peak of the diabetic wave are calculated from the maximum absolute value of the first set of concave points and the maximum absolute value of the convex points of the main peak. The diabetic pre-wave and diabetic wave are then merged. For Class I pulse waves, calculate the position of the main peak, and merge the diabetic wave and diabetic pre-wave with the main peak; Among them, the method for calculating the peak and trough points in the type A pulse wave is the extreme value method; And / or, in the type B pulse wave, the peak and trough points are calculated using the extreme value method, and the single-point decreasing convex and concave points are calculated using the curvature method; And / or, in the C-type pulse wave, the peak and trough points are calculated using the extreme value method; And / or, in the type D pulse wave, the peak and trough points are calculated using the extreme value method, and the single-point decreasing convex and concave points are calculated using the curvature method; And / or, in the E-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the F-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the G-type pulse wave, the peak and trough points are calculated using the extreme value method; And / or, in the H-type pulse wave, the method for calculating the peak and trough points is the curvature method; And / or, in the Type I pulse wave, the method for calculating the peak and trough points is the extreme value method.

3. The pulse wave time-domain feature classification system according to claim 2, characterized in that, The calculation method for the time-domain characteristic parameters of pulse waves from Type A to Type I is as follows: The calculation method of time domain characteristic parameters of Class A pulse wave: (1) Calculate the coordinates of all peak points and valley points of the pulse wave by means of the pulse wave extreme value method; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the position of the valley point to the left of each peak point except the main wave peak, form a peak-valley pair, and calculate the height difference between the peak and valley; (4) Determine the position of the diabetic pre-wave and diabetic wave: form a pair of peak-valley pairs by forming adjacent valleys and peaks, and define the two points with the largest distance between the peak and valley pairs as the diabetic pre-wave and diabetic wave from left to right according to their coordinate positions; The calculation method of time-domain characteristic parameters of type B pulse wave is as follows: (1) Calculate all peak points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing points by the curvature method; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is obtained; the horizontal coordinate position corresponding to the peak point is x c Let the two points corresponding to the decrease of 1 / 3 in the peak value of the main wave be a and b, and let the x-coordinate of point b be x. b Within the sampling point threshold range (x) c， x b ), and define the point with the largest difference between the peak and the valley as the coordinate of the diabetic pre-wave; (5) Determine the diabetic wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a convex and concave single-point decreasing convex and concave point pair, and calculate the height difference between the convex and concave single-point decreasing convex and concave point pairs, e is the peak of the diabetic pre-wave, and the horizontal coordinate corresponding to point e is x. e Within the threshold range of sampling points (x) e， (0.7N), the point with the largest difference is defined as the coordinate of the diphtheria wave; N is the total number of sampling points; The calculation method of time-domain characteristic parameters of C-type pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (3) Determine the diabetic anterior wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. The sampling point threshold range is 1-0.7N. The point with the largest difference between the peak and valley is defined as the coordinate of the diabetic anterior wave; N is the total number of sampling points. The calculation method of time-domain characteristic parameters of type D pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is obtained. e is the peak of the diabetic wave, and the horizontal coordinate of point e is x. e Within the threshold range of sampling points (x) e， 0.7N), the point with the largest difference between the peak and valley is defined as the coordinate of the diabetic wave; N is the total number of sampling points; (5) Determine the diabetic wave: the position of the concave single point decreasing convex and concave point to the left of each convex single point decreasing convex and concave point constitutes a pair of convex and concave single points decreasing convex and concave points, and the height difference between the pairs of convex and concave single points decreasing convex and concave points is calculated; within the range of the horizontal coordinates of the diabetic wave valley and the main wave peak, the point with the largest difference is defined as the coordinate of the diabetic wave; The calculation method of time domain characteristic parameters of E-type pulse wave: (1) Calculate the coordinates of all peak points and valley points of pulse wave through the extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point within the range to the right of the main wave peak, form a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; (5) Determine the position of diabetic pre-wave and diabetic wave: define the two points with the largest height difference of convex and concave single-point decreasing convex and concave points from left to right according to their coordinate positions as diabetic pre-wave and diabetic wave; The calculation method of time-domain characteristic parameters of type F pulse wave is as follows: (1) Calculate the coordinates of all peak points and valley points of the pulse wave through the extreme points of the pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of the pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic pre-wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of the diabetic wave and the main wave peak, the point with the largest difference is defined as the coordinate of the diabetic pre-wave; The calculation method of time-domain characteristic parameters of type G pulse wave: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinate of the largest peak point as the position of the main wave peak; (3) Determine the diabetic wave: the position of the peak and valley points to the left of each peak point constitutes a peak-valley pair, and the height difference between the peak and valley is calculated. e is the peak of the diabetic wave, and the horizontal coordinate of point e is x. e Within the threshold range of sampling points (x) e， 0.7N), the point with the largest difference between the peak and the valley is defined as the coordinate of the diphtheria wave; N is the total number of sampling points; The calculation method of time-domain characteristic parameters of type H pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Calculate the position coordinates of all convex and concave single-point decreasing convex and concave points through the calculation method of single-point decreasing convex and concave points of pulse wave; (3) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak; (4) Determine the diabetic wave: the position of the concave single-point decreasing convex and concave point to the left of each convex single-point decreasing convex and concave point forms a pair of convex and concave single-point decreasing convex and concave points, and calculate the height difference between the pairs of convex and concave single-point decreasing convex and concave points; within the range of the horizontal coordinates of diabetic wave and main wave peak, the point with the largest difference is defined as the coordinate of diabetic wave; The calculation method of time-domain characteristic parameters of type I pulse wave is as follows: (1) Calculate all peak points of pulse wave through extreme points of pulse wave; (2) Determine the main wave: take the coordinates of the largest peak point as the position of the main wave peak.

4. A pulse wave time-domain feature recognition system, characterized in that, include: The input module is configured to input a pulse wave; The prediction module integrates the pulse wave time-domain feature classification system according to any one of claims 1-3, wherein the classification module is configured to identify the pulse wave through a machine learning model within the threshold range of the sampling points, classify the dicrotic anterior wave into three features: extreme point, single-point decreasing convex and concave point, and feature fusion, classify the dicrotic wave into three features: extreme point, single-point decreasing convex and concave point, and feature fusion, and classify the pulse wave into class A to class I pulse waves based on the classification results of the dicrotic anterior wave and dicrotic wave; The output module is configured to output the classification result of the prediction module.

5. The pulse wave time-domain feature recognition system according to claim 4, characterized in that, The pulse wave is the single-cycle optimal pulse wave; the three characteristics of the dicrotic pre-wave and the three characteristics of the dicrotic wave are calculated using the single-cycle optimal pulse wave.

6. The pulse wave time-domain feature recognition system according to claim 4, characterized in that, The machine learning model is selected from the convolutional neural network feature recognition model.

7. The pulse wave time-domain feature recognition system according to claim 6, characterized in that, The convolutional neural network feature recognition model includes a convolutional neural network feature recognition model for the dicrotic anterior wave and a convolutional neural network feature recognition model for the dicrotic wave. The two models are used to identify the types of the dicrotic anterior wave and the dicrotic wave of the pulse wave, respectively, and determine that the pulse wave is one of the nine types of waveforms from type A to type I.

8. The pulse wave time-domain feature recognition system according to claim 4, characterized in that, The prediction module is also configured to calculate the temporal characteristic parameters of the pulse wave based on the classification results of the pulse wave.

9. A computer-readable storage medium, characterized in that, It stores: a computer program for implementing the pulse wave time-domain feature classification system according to any one of claims 1-3, or a computer program for implementing the pulse wave time-domain feature recognition system according to any one of claims 5-9.

Images