Leather pulse wave identification method

Abstract

The invention relates to signal processing, in particular to a leather pulse condition signal recognition processing method which comprises the steps that pulse waves are continuously collected through a pulse diagnosis device, and original pulse condition data in rated time are obtained; the original pulse condition data within the rated time are periodically collected, the original pulse condition data within the rated time are divided into n continuous original pulse condition data, and n is an integer larger than or equal to 2; performing low-pass filtering processing on each continuous original pulse condition data, eliminating noise, and obtaining n pieces of de-noised pulse condition data; effective feature extraction of pulse waves is carried out on each piece of denoised pulse condition data, a pulse recognition model is input to carry out preliminary feature matching recognition screening processing, and preliminary matching recognition data is obtained; and carrying out effective feature extraction of pulse waves on the preliminary matching identification data, inputting the preliminary matching identification data into a leather vein identification model to carry out final feature matching identification screening to obtain final matching identification data, and then outputting the final matching identification data as an identification result.

Description

TECHNICAL FIELD

[0001] The present application relates to signal processing, and particularly relates to a recognition processing method of a leather pulse pulse condition signal. BACKGROUND

[0002] Leather pulse is a type of pulse condition in traditional Chinese medicine, characterized by floating and beating, hollow and hard, like pressing a drum skin. In traditional Chinese medicine pulse diagnosis, leather pulse is considered to be closely related to the occurrence and development of certain diseases, so accurate identification and analysis of leather pulse is of great significance. Current pulse diagnosis technology has some challenges in the identification of leather pulse. Traditional pulse condition diagnosis mainly relies on the subjective experience of doctors, lacks objective theoretical basis and standardized diagnosis method, resulting in certain subjectivity and uncertainty in the identification result of leather pulse.

[0003] In the past few years, with the rapid development of new material technology and electronic computer technology, these technologies have been widely applied in the medical field. In the acquisition, processing and visualization of pulse information, remarkable progress has been made, providing a foundation for the establishment of the theoretical basis of pulse signal. These breakthroughs in new material technology make pulse diagnosis instruments and other pulse collection devices more accurate and efficient. By collecting the pulse data of patients and combining advanced signal processing algorithms, key feature information in the pulse can be accurately extracted. At the same time, the continuous development of electronic computer technology provides powerful computing power and algorithm support for big data analysis and machine learning. The patent with the authorization publication number CN116089797A uses convolutional neural networks to simulate the characteristics of human neuron structure and layer-by-layer information transmission to realize automatic feature extraction and ultimately realize pulse recognition. The patent with the authorization publication number CN113712516A discloses a method for establishing a traditional Chinese medicine pulse recognition model based on a time series convolution network. The pulse recognition method proposed by the invention has high accuracy and can provide reference for the diagnosis of the health status of the human cardiovascular system. In the paper "Pulse Signal Recognition Research Based on Time Series and Time Series Convolution Network" by Zhu Guangyao et al., the pulse signal is preprocessed and sequence normalized to obtain pulse signal time series of consistent length. The deep learning network based on time series convolution is used to extract the features of the sequence form, and a pulse signal recognition model is established. In the paper "Pulse Signal Recognition Research Based on Weighted Soft Voting Fusion Model" by Liu Qichao et al., the boundary synthesis minority class sample oversampling technology is used to solve the data imbalance problem, and a weighted soft voting fusion model based on XGBoost, random forest, LightGBM, and gradient boosting decision tree is constructed. The final model outputs the specific pulse category, which has lower operation complexity and higher accuracy compared to common pulse recognition methods, and the shorter training time also makes it more clinically valuable in various pulse signal recognition. Some patents also use different deep learning models and methods to recognize pulse. The above inventions can better identify pulse and have high accuracy, but deep learning requires high hardware costs and is complex to operate. The patent with the authorization publication number CN119785982A discloses a method based on model training for identifying the processing method of the leather pulse wave. The invention has low cost and simple operation, but in actual operation, it is found that the invention cannot well distinguish the actual pulse signal of the leather pulse wave, so on this basis, the invention introduces the unique characteristics of the leather pulse wave for recognition, and after being tested by many vice-high doctors, it has high accuracy in pulse differentiation. SUMMARY

[0004] The technical problem to be solved by the present invention is how to accurately identify the leather pulse wave in the data collected by the pulse diagnosis device.

[0005] The technical scheme adopted by the present application is as follows: a pulse wave recognition method is performed according to the following steps:

[0006] Step one, continuously collecting pulse wave by using a pulse diagnosis device to obtain original pulse condition data of a rated time;

[0007] Step two, periodically collecting the original pulse condition data of the rated time, dividing the original pulse condition data of the rated time into n continuous original pulse condition data, n being an integer greater than or equal to 2;

[0008] Step three, performing low-pass filtering processing on each continuous original pulse condition data to eliminate noise and obtain n de-noised pulse condition data;

[0009] Step four, performing effective feature extraction on each de-noised pulse condition data, inputting the pulse wave into a pulse recognition model to perform preliminary feature matching recognition screening processing, and obtaining preliminary matching recognition data;

[0010] Step five, performing effective feature extraction on the preliminary matching recognition data, inputting the pulse wave into a pulse recognition model to perform final feature matching recognition screening, obtaining final matching recognition data, and then outputting the final matching recognition data as a recognition result. The rated time is 20-30 seconds, the sampling period is 5 seconds when periodically collecting the original pulse condition data of the rated time, and n is 4-6.

[0011] The pulse diagnosis device is a pulse diagnosis instrument, and the collected part is the pulse amplitude data of the inch part, the joint part or the foot part. The effective features include the main wave width, the main wave height, the tidal wave height, the heavy pulse wave height, the descending middle isthmus height, the main peak angle and the ascending branch angle.

[0012] When performing the feature matching recognition screening processing, the matched features include the pulsation starting point (X1, Y1), the main wave peak point (X2, Y2), the main wave ending point (X3, Y3) and the pulsation ending point (X4, Y4) of the pulse wave signal in a single pulse wave period. In step four, when performing the preliminary feature matching recognition screening processing in the pulse recognition model, the pulse wave that meets the following screening processing is the pulse wave:

[0013] The ascending branch slope k1 of the pulse wave signal, Remove the pulse wave signal with k1<11.43;

[0014] The height of the main wave peak of the pulse wave signal is Y2, the descending middle isthmus height is Y3, and the pulse wave signal with Y2>Y3 is removed;

[0015] The systolic period (ejection period) time t of the pulse wave signal is t=(X4-X1)*0.008, and the pulse wave signal with t>0.72 seconds is removed.

[0016] ​In step five, when the pulse recognition model performs final feature matching recognition screening, the pulse wave that meets the following screening processing is the pulse wave: the falling slope k2 of the main wave peak point and the fluctuation end point of the pulse wave signal, Remove the pulse wave signal with k2>-9.56;

[0017] The distance L1 from the main wave end point to the rising branch of the pulse wave in the pulse wave signal,

[0018] The distance L2 from the main wave end point to the main wave peak point,

[0019] The main wave angle θ, Remove the pulse signal that does not meet θ>33°.

[0020] In step four, the pulse wave signal contains the floating pulse feature, the angle of the main wave rising branch is greater than 84°, and the rising branch has no small wave, and the ratio of the height of the falling branch to the height of the main wave is less than 0.15; the pulse wave signal only has the main wave and the heavy wave, and no tidal wave (and not judged as a string pulse pulse wave) is the said Ji pulse wave.

[0021] The data points of the pulse wave original data are composed of multiple data points, and the period extraction labels the pulse wave original data as multiple periods, that is, the pulse wave original data is divided into multiple repeated periods.

[0022] The effective feature definitions of the application are from “Clinical Illustration of Chinese Pulse Diagnosis” (Peng Qinghua, Xie Mengzhou, Clinical Illustration of Chinese Pulse Diagnosis [M]. Beijing: Chemical Industry Press, 2018.).

[0023] The definition of the beat starting point is: in a single pulse wave cycle, the first time point where the signal starts to rise continuously from the baseline level and the increment of the continuous 3-5 sampling points is positive before the main wave peak point, which is the mark of the start of the beat of this pulse cycle. The definition of the main wave peak point is: in a single pulse wave cycle, the time point where the signal amplitude reaches the global maximum value, which is the most significant feature point in the cycle and is used as a reference for other feature points. The definition of the main wave end point is: in a single pulse wave cycle, after the main wave peak point, the first relative trough point appears in the falling process of the pulse wave signal, which marks the end of the main wave stage; the definition of the beat end point is: in a single pulse wave cycle, after the main wave end point, the pulse wave signal fluctuates, and then falls to a stable state close to the baseline or prepares for the next cycle beat, which marks the complete closed loop of this pulse cycle.

[0024] The beneficial effects of the present application are: the present application adopts "pulse wave preliminary screening → leather pulse fine identification", which is more simplified and efficient compared with the complex deep learning method. And it solves the problem of difficult to distinguish between leather pulse and leather pulse waveform, although a simple identification step is adopted, but the method can still maintain a relatively high identification accuracy, compared with the deep learning method, the implementation of the method is more simple, the cost is lower, therefore it is more feasible and practical. The present application has the advantages of simplicity, high accuracy and low cost, and is expected to be widely used in the field of pulse diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is the whole pulse signal recognition flow chart;

[0026] Figure 2 is the schematic diagram of the leather pulse pulse image of the present application for a single cycle;

[0027] Figure 3 is the pulse image identified as leather pulse;

[0028] Figure 4 is the pulse image identified as non-leather pulse. DETAILED DESCRIPTION

[0029] The present application will be specifically described below in combination with the drawings and specific embodiments, so that the present application can be better understood and implemented by the skilled in the art.

[0030] Example 1

[0031] The present application provides a leather pulse pulse image recognition method. The specific identification steps are as follows:

[0032] Obtain the pulse image original signal of the target. In order to identify the pulse image, the pulse image original signal of the target person is needed as the basic data. In order to ensure the reliability of the data, the present application adopts a high sensitivity acquisition device, and the pulse wave data of the target person's left hand cun, guan and chi under light, medium and heavy pressure is collected. The sampling period under each part is 6 seconds, and a total of 18 groups of data are collected.

[0033] Filtering and noise reduction processing. The collected periodic signal is subjected to low-pass filtering processing, and a low-pass filter is used to reduce the noise of the pulse wave signal, so as to remove the noise components in the original signal.

[0034] Step three: using algorithm to extract the period of filtered pulse. Since the sampling period of pulse wave signal is 5 seconds, and the heart rate of human is generally in the range of 60-100 times per minute, the extracted pulse wave pulsation period signal is usually 4-6 periods. The period-divided pulse signal is input into the model for recognition. The filtered and period-extracted pulse signal is input into the mathematical model of pulse wave recognition method for recognition, and the signal conforming to the pulse wave is input into the mathematical model of leather pulse for recognition, which specifically includes extracting effective features and performing feature matching.

[0035] To clarify the related description of feature matching, the following is stated first. The pulsation starting point, main wave peak point, main wave ending point, and pulsation ending point of the pulse wave signal are set as (X1, Y1), (X2, Y2), (X3, Y3), and (X4, Y4) in sequence.

[0036] First, recognize the pulse wave

[0037] First, calculate the rising slope k1 of the retained pulse wave signal, remove the pulse wave signal with k1 < 11.43; judge the height Y2 of the main wave peak and the height Y3 of the descending middle isthmus, remove the pulse wave signal with according to the range set by the A / D circuit of the pulse diagnosis instrument.

[0038] Calculate the systolic period (ejection period) time t of the retained pulse wave signal, t = (X4-X1)*0.008, remove the pulse wave signal with t > 0.72 seconds.

[0039] Next, recognize the leather pulse

[0040] First, calculate the descending slope k2 of the main wave peak point and the wave ending point of the retained pulse wave signal, remove the pulse wave signal with k2 >-9.56;

[0041] Second, calculate the ratio of the ordinates of the main wave peak point and the main wave ending point, remove the filtered signal that does not satisfy .

[0042] Then, calculate the distance L1 from the main wave ending point to the rising branch of the retained pulse wave signal, the distance L2 from the main wave ending point to the main wave peak point, the main wave angle θ, remove the pulse signal that does not satisfy θ > 33°.

[0043] Finally, after the above multiple screening, the angle of the main wave rising branch is greater than 84°, and the rising branch has no small wave, the ratio of the height of the descending branch to the height of the main wave is less than 0.15, the waveform has only the main wave and the heavy wave, and there is no tide wave (and it is judged not to be a string pulse) is the leather pulse wave. The whole data recognition process is accurate and can accurately identify the leather pulse wave, which prepares for the next step.

[0044] The result is determined as follows: if the proportion of the cycle identified as the leather pulse wave to the total extracted cycle is >80% (for example, 4 of the 5 extracted cycles meet the above conditions), combined with Figure 3 The overall characteristics of the leather pulse diagram are determined as the leather pulse.

[0045] The identification result is output. According to the model, the input pulse signal is analyzed and judged, and the result of pulse identification is output, that is, the conclusion of classifying and identifying the pulse wave of the target person.

[0046] Example 2

[0047] In this embodiment, we will show how to apply the pulse identification method of the present application to identify and exclude the pulse wave that meets the leather pulse wave but does not meet the typical characteristics of the leather pulse wave.

[0048] First, the pulse diagnosis device is still used to collect the pulse wave raw data of the object to be tested at the inch, Guan, and chi positions, and the pulse wave is also divided into multiple cycle signal segments through the cycle extraction process.

[0049] After low-pass filtering the cycle signal, attention is paid to the pulse wave that does not meet the key characteristics of the leather pulse wave. For such signals, the following identification steps are performed:

[0050] The descending branch slope k2 of the main wave peak point and the end point of the wave is calculated, and if k2>-9.56, it indicates that the descending branch decays slowly and lacks the rapid drop characteristics of the leather pulse "outer hardness". Through the above screening, it can be initially identified as a non-leather pulse wave signal.

[0051] The ratio of the main wave peak point and the main wave end point ordinate is calculated If the ratio is not within the range of 1.4-1.65, it does not meet the amplitude characteristics of the leather pulse "hollow outer hardness".

[0052] The main wave angle θ is calculated If θ<33°, the waveform is flat and has no leather pulse "tightness";

[0053] After the above screening, such signals are determined as non-leather pulse waves.

[0054] It is apparent that the above examples are only illustrative and do not limit the embodiments. Those skilled in the art can make other forms of changes or adjustments on the basis of the above description. It is not necessary to list all the embodiments, and the obvious changes derived therefrom are still within the protection scope of the present application.

Claims

1. A method for identifying pulse waves, characterized in that, Follow these steps: Step 1: Use a pulse diagnosis device to continuously collect pulse waves and obtain raw pulse data for a specified time. Step 2: Periodically collect the raw pulse data for the rated time, and divide the raw pulse data for the rated time into n consecutive raw pulse data, where n is an integer greater than or equal to 2. Step 3: Perform low-pass filtering on each consecutive raw pulse data to eliminate noise and obtain n denoised pulse data. Step 4: Extract effective features of hollow pulse waves from each denoised pulse data, and input them into the hollow pulse recognition model for preliminary feature matching and screening to obtain preliminary matching and recognition data; Step 5: Extract effective features of the hollow pulse wave from the preliminary matching and identification data, and input it into the leather pulse identification model for final feature matching and identification screening to obtain the final matching and identification data. Then, output the final matching and identification data as the identification result.

2. The method for identifying pulse waves according to claim 1, characterized in that: The rated time is 20-30 seconds. When periodically collecting the raw pulse data for the rated time, the sampling period is 5 seconds and n is 4-6.

3. The method for identifying pulse waves according to claim 1, characterized in that: The pulse diagnosis device is a pulse diagnosis instrument. When collecting pulse waves, the pulse amplitude data is collected from the cun, guan, or chi positions.

4. The method for identifying pulse waves according to claim 1, characterized in that: Effective features include main wave width, main wave height, tidal wave height, diurnal wave height, descending mid-straight line height, main peak angle, and ascending branch angle.

5. The method for identifying pulse waves according to claim 4, characterized in that: During the feature matching and screening process, the matched features include the pulse wave signal's pulsation start point (X1, Y1), main wave peak point (X2, Y2), main wave end point (X3, Y3), and pulsation end point (X4, Y4) within a single pulse wave cycle. In step four, during the preliminary feature matching and screening process in the hollow pulse identification model, the pulse wave that meets the following screening criteria is considered a hollow pulse: The slope k1 of the rising limb of the pulse wave signal Remove pulse wave signals where k1 < 11.43; The height of the main peak of the pulse wave signal is Y2, and the height of the septum is Y3. (After removing...) The pulse wave signal; The systolic (ejection) time t of the pulse wave signal is t = (X4 - X1) * 0.

008. Pulse wave signals with t > 0.72 seconds are removed.

6. The method for identifying pulse waves according to claim 5, characterized in that: In step five, during the final feature matching and screening process of the leather vein recognition model, those that meet the following screening criteria are considered leather vein waves: The slope k2 of the descending branch at the peak of the main wave and the end of the wave in the pulse wave signal. Remove pulse wave signals where k2 > -9.56; The distance L1 from the end of the main wave to the rising limb of the pulse wave in the pulse wave signal. The distance L2 from the end point of the main wave to the crest point of the main wave. The angle θ between the main waves Remove pulse signals that do not satisfy θ>33°.

7. The method for identifying pulse waves according to claim 1, characterized in that: In step four, the pulse wave signal contains floating pulse characteristics, the angle of the rising branch of the main wave is greater than 84°, and there are no small waves in the rising branch. The ratio of the height of the descending mid-wave to the height of the main wave is less than 0.

15. The pulse wave signal only has the main wave and the diurnal wave, and there is no tidal wave (and it is determined not to be a string-like pulse wave), which is the hollow pulse wave.

Images