CN114648040A

CN114648040A - Method for extracting and fusing multiple physiological signals of vital signs

Info

Publication number: CN114648040A
Application number: CN202210188212.1A
Authority: CN
Inventors: 韩宏光
Original assignee: Yujingquan
Current assignee: Yujingquan
Priority date: 2022-02-28
Filing date: 2022-02-28
Publication date: 2022-06-21

Abstract

The invention discloses a vital sign multi-physiological signal extraction and fusion analysis method, which is characterized by extracting the characteristics of electrocardio, electroencephalogram, myoelectricity, pulse, cardiopulmonary sound and other parameter information collected by equipment based on artificial intelligence, performing multi-physiological signal fusion analysis by using an SVM (support vector machine), automatically obtaining the physiological state of a patient, and quickly obtaining feedback and response by medical staff. The computer is used for automatically analyzing the data and judging the physiological state of the wounded patient, so that the treatment time is greatly saved, and the rescue rate of the wounded patient is improved.

Description

Method for extracting, fusing and analyzing multiple physiological signals of vital signs

Technical Field

The invention belongs to the technical field of vital sign detection, and particularly relates to a vital sign multi-physiological signal extraction and fusion analysis method.

Background

With the rapid development of industry and transportation industry, the probability of natural disasters increases, so that various emergencies occur continuously, a large number of sick and wounded persons can be generated in a short time, and the sick and wounded persons are seriously ill and complicated. Therefore, the emergency medical post-treatment work for researching the sick and wounded has important practical significance. After the emergency medical treatment is performed on the wounded, multiple physiological signals of vital signs of the wounded are monitored in real time, such as electrocardio, electroencephalogram, myoelectricity, pulse, cardiopulmonary sound and the like, but the existing medical facilities can only monitor various physiological signals of the wounded by using various monitoring instruments, and the physiological state of the wounded is obtained after data analysis and judgment of medical staff, so that the mode is low in efficiency and not beneficial to treatment of a large number of wounded in emergency.

Disclosure of Invention

In order to solve the problems, the invention designs a vital sign multi-physiological signal extraction and fusion analysis method which can be applied to vital sign monitoring equipment, realizes the rapid analysis and processing of monitoring data, and feeds back the physiological state of a patient to medical staff in time, thereby greatly improving the rescue and treatment rate of the patient.

In order to achieve the purpose, the invention adopts the following technical scheme: the method comprises the following steps:

1) extracting electrocardio, electroencephalogram, myoelectricity, pulse and heart-lung sound signals;

2) preprocessing each signal and extracting characteristics in the step 1);

3) performing fusion analysis on multiple physiological signals by the SVM; wherein the content of the first and second substances,

the SVM performs a multi-physiological signal fusion analysis method, and selects each physiological signal characteristic in the step 2) by using a correlation coefficient method, wherein the method comprises the following steps:

a. calculating the correlation coefficient of each feature and the training label in the training set;

b. selecting corresponding characteristics with correlation coefficients higher than 0.1, inputting the characteristics into each SVM classifier for training, respectively calculating the accuracy of a training set of each characteristic in the training process, and setting the accuracy of the training set with the training accuracy lower than 50% in the formula to zero;

c. and obtaining a final classification result by using a voting method.

The method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: the preprocessing and feature extraction of the electrocardiosignals comprise electrocardiosignal denoising preprocessing, high-quality electrocardiosignal screening and electrocardiosignal feature extraction, wherein the electrocardiosignal denoising preprocessing adopts a soft threshold value method to process low-scale wavelet coefficients d1 and d2, and high-frequency noise is eliminated through a large-amplitude attenuation coefficient method; and filtering the scale 3 wavelet coefficient containing the important input signal by adopting a soft and hard threshold compromise algorithm to obtain a clean electrocardiosignal.

The method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: extracting the characteristics of the electrocardiosignals, and finishing the detection of singular points of the ECG signals based on a biorthogonal quadratic B-spline wavelet transform method; and the positioning precision of the R wave is improved by adopting a dynamic valve value method. Analyzing the wavelet coefficient D4, extracting an extreme value pair, obtaining the extreme value pair for accurately positioning the R wave by adopting threshold processing and filtering, restoring the position of the extreme value pair to the reconstructed ECG signal, and finding the maximum value in the maximum value interval to obtain R wave positioning; after the R wave is positioned, Q, S waves can be positioned by searching a minimum value point near the R wave according to the position of the R wave; the length of a detection interval is controlled by taking the positions of two adjacent R waves as a marker post and the mean value of an RR interval, so that two heart beats between the two adjacent R waves, respective T waves and P waves are accurately detected; and measuring respective detection areas of the T wave and the P wave to complete accurate positioning of P, Q, R, S, T waves of the ECG signal.

The method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: extracting the basic features and the characteristics of the electroencephalogram signals, namely extracting the multiple features of the electroencephalogram signals based on wavelet packet transformation; determining signal frequency of the obtained electroencephalogram signal, calculating a decomposition scale j, obtaining a wavelet packet sub-band, reconstructing and obtaining 5 rhythm waves, calculating fast and slow wave energy and sample entropy, and evaluating the wave energy and the sample entropy to select a feature vector meeting requirements.

The method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: extracting an iEMG value and an RMS value from the electromyographic signals to be used as electromyographic time domain characteristics; and carrying out fast Fourier transform on the sEMG signal to obtain the electromyogram frequency domain characteristics.

The method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: removing burrs and baseline drift of the pulse signals by using wavelet transformation, and extracting peak points of the pulse waves by using a differential dual-threshold method; taking the area change of the pulse wave chart as the pulse wave waveform characteristic quantity K value;

the method for extracting and fusing the vital sign multi-physiological signals is further improved as follows: the extraction of the heart-lung sound signals comprises the following steps;

a) carrying out short-time Fourier transform on the collected heart-lung sound original signals to obtain original signal frequency spectrums;

b) decomposing the frequency spectrum signal into two sub non-negative matrixes W and H by using a non-negative matrix decomposition NMF technology;

c) and (3) carrying out classification training on each basis vector of the basis matrix W and the weight vector of the coefficient matrix H by using a clustering algorithm, and dividing the basis vectors into two types of signals, namely heart sound spectrum signals and lung sound spectrum signals:

d) the heart sound spectrum and lung sound spectrum signals on the time frequency domain are converted into waveform signals on the time domain by using an inverse short-time Fourier transform (ISTFT) technology.

In order to achieve the purpose, the invention adopts the following technical scheme: the device comprises a signal measuring unit, a data processing unit, a data transmission unit and a power management unit; the signal acquisition unit acquires signals of electroencephalogram, electrocardio and myoelectricity through electrodes, acquires blood pressure through a cuff with a high-precision pressure sensor, measures respiration by utilizing a chest elastic band and acquires body temperature by adopting a contact type body temperature measurement sensor;

the method for extracting and fusing the multiple physiological signals of the vital signs is further improved by the wireless vital sign monitoring equipment applied as the method for extracting and fusing the multiple physiological signals of the vital signs: the data processing unit preprocesses the acquired electroencephalogram, electrocardio, heart sound and electromyogram signals through a signal filtering unit and an external ADC signal conversion unit, wherein the signal filtering unit simply filters some interference signals in the original signals acquired by the signal measuring unit; the signals are processed by the interface protection circuit and the signal filter circuit and then respectively transmitted to the external high-precision special ADC and the ADC integrated in the CPU of the equipment for data conversion from analog quantity to data quantity of the signals. And the data is transmitted to a CPU of the equipment after ADC data conversion, and then the CPU of the equipment uploads the data to a host through controlling a low-power wireless transmission module for the host to analyze and process the data.

The method for extracting, fusing and analyzing the multiple physiological signals of the vital signs can be applied to wearable wireless vital sign monitoring equipment, and the vital sign signals of the patient acquired by the wearable wireless vital sign monitoring equipment are quickly analyzed and processed to obtain the physiological state of the patient and fed back to medical staff, so that the medical staff can quickly judge and ask for help, the treatment time is greatly saved, and the treatment rate of the patient is improved.

Drawings

FIG. 1 is a flow chart of multi-signal fusion determination according to the present invention;

FIG. 2 is a diagram of classification of brain electrical signals in accordance with the present invention;

FIG. 3 is a flow chart of electroencephalogram feature extraction in the present invention;

FIG. 4 is a waveform diagram of pulse waves in different physiological and pathological states;

FIG. 5 is a flow chart of the cardiopulmonary sound signal separation according to the present invention;

fig. 6 is a flow chart of fusion analysis of multiple physiological signals in the present invention.

Detailed Description

The present invention is further described with reference to the following drawings and specific examples, which are provided for the purpose of explaining the technical solutions of the present invention in detail.

The vital sign multi-physiological signal extraction and fusion analysis method is applied to a central control system of wearable whole-course vital sign wireless monitoring equipment, and is used for automatically processing and analyzing acquired data information and obtaining the physiological state of a wounded patient. The wearable whole-course vital sign wireless monitoring equipment comprises a signal measuring unit, a data processing unit, a data transmission unit and a power management unit.

Specifically, the signal acquisition unit comprises the steps of acquiring electroencephalogram, electrocardio and myoelectricity signals through electrodes, acquiring blood pressure through a cuff with a high-precision pressure sensor, measuring respiration through a chest elastic band and acquiring body temperature through a contact type body temperature measuring sensor.

The data processing unit preprocesses the acquired electroencephalogram, electrocardio, heart sound and electromyogram signals through a signal filtering unit and an external ADC signal conversion unit, wherein the signal filtering unit simply filters interference signals in the original signals acquired by the signal measuring unit; the signals are processed by the interface protection circuit and the signal filter circuit and then respectively transmitted to the external high-precision special ADC and the ADC integrated in the CPU of the equipment for data conversion from analog quantity to data quantity of the signals. And the data is transmitted to a CPU of the equipment after ADC data conversion, and then the CPU of the equipment uploads the data to a host through controlling a low-power wireless transmission module for the host to analyze and process the data.

Furthermore, the data transmission unit transmits original data of the data acquisition unit to the equipment host without loss and transmits data returned by the host to the equipment, and is based on a low-power wireless transmission technology, so that the defects of inconvenience in wired transmission operation, large occupied space, high power consumption and the like are overcome, the limitation on application scenes is reduced, the data transmission unit is based on the low-power wireless technology, the power consumption of the equipment is greatly reduced on the premise of not reducing the data transmission rate, and the service life of the equipment is prolonged; the unit realizes automatic identification and binding with a host in use, is convenient to operate, encrypts and transmits data, and provides system safety.

Further power management unit, equipment adopt the battery power supply, have greatly improved the flexibility that equipment used, based on low-power consumption power management scheme, extension equipment standby and live time, when battery electric quantity crosses low automatic alarm reminds to charge, under the charged state, the automatic switch-over of equipment realization power, under the prerequisite that does not influence equipment use, the realization is to the charging of battery. When the equipment is in a standby state, the power management unit implements low-power-consumption operation management on the equipment power supply, and the endurance time of the equipment is greatly prolonged.

The feature extraction of the electrocardiosignal needs to carry out denoising processing on the electrocardioparameters obtained on a patient body. The electrocardiosignal noise mainly comprises baseline drift, power frequency interference and electromyographic noise, and the frequency band distribution of each noise signal contained in the electrocardiosignal ECG is also different: the baseline drift is 0-0.5Hz, and the power frequency interference is 50-60 Hz; myoelectric noise is 50-2 KHz. Furthermore, the central frequency of the QRS wave group, the main component of the ECG signal, is about 17 Hz. QRS wave group energy is mainly distributed on the scales 3 and 4 after wavelet transformation, power frequency interference energy is mainly distributed on the scale 2 after wavelet transformation, and myoelectricity noise energy is mainly distributed on the scales 1, 2 and 3 after wavelet transformation. Therefore, the low-scale wavelet coefficients d1 and d2 are processed by a soft threshold value method, and high-frequency noise is eliminated by a large-amplitude attenuation coefficient method. Aiming at the scale 3 wavelet coefficient containing important input signals, the algorithm of soft and hard threshold compromise is adopted, so that the electromyographic noise is eliminated, the original input signals are kept as far as possible, and the accurate extraction of subsequent electrocardio characteristic parameters is not influenced. At the same time, baseline wander is suppressed by the all-pass down-pass filter.

The quality of the denoised electrocardiosignals is judged, the low-quality electrocardiosignals are screened out, and the evaluation standard is as follows: a1: the threshold is greater than 40% by the 3mv portion, which is considered to meet the criterion.

A2: the criterion is satisfied when the first derivative is greater than 0.3 (i.e., the number of peak portion samples) is greater than 40%.

A3-greater than 80% of the lead fall-off, this criterion is met.

A4: the portion satisfying the determination condition in a1, a2, A3 is a potential failing point, and when the potential failing portion is larger than 68.5%, the criterion is satisfied.

Through the four judgment standards, signals with poor contact, lead falling and poor quality can be screened out without processing, and signals with good quality can be used for disease analysis.

And (4) performing feature extraction on the high-quality electrocardiosignals obtained after screening. The invention adopts a method based on biorthogonal quadratic B-spline wavelet transform to complete the detection of the singular points of the ECG signal. The ECG signal R wave presents an extreme value pair form of the maximum amplitude value under the scale 4, the frequency band range (11.3-22.5 Hz) of the scale 4 is closest to the QRS wave central frequency (17 Hz), and the influence of unfiltered and clean noise outside the bandwidth is effectively avoided, so that the detection precision of the R wave is ensured. And the positioning precision of the R wave is improved by adopting a dynamic valve value method. And analyzing the wavelet coefficient D4, extracting an extreme value pair, obtaining the extreme value pair for accurately positioning the R wave by threshold processing and filtering, restoring the position of the extreme value pair to the reconstructed ECG signal (with time shift), and finding the maximum value in the maximum value interval to obtain the R wave positioning. After the R wave is positioned, Q, S waves can be positioned by finding a minimum value point near the R wave according to the position of the R wave.

In the detection of the P wave and the T wave, the length of a detection interval is controlled by taking the positions of two adjacent R waves as a benchmark and the average value of RR intervals, so that two heartbeats between the two adjacent R waves are accurately detected. A respective T-wave and P-wave. The respective probe regions of the measurement T wave and the measurement P wave are such that the P, Q, R, S, T waves of the ECG signal are precisely located, providing a data source for later model construction.

The extraction of the characteristics of the brain electrical signal, which is the total reflection of the brain central nervous system to the change of human physiological activity and is formed by the synchronous electrical activity of related neuron cell groups in the cerebral cortex. Normally, the transmembrane resting potential of a neuronal cell is about-60 mV to-70 mV, and when the cell is externally stimulated, its internal and external environment changes, causing changes in the transmembrane potential of the neuronal cell, which changes are transmitted between different neurons via a structure known as synapse, resulting in the formation of bioelectrical currents. The bioelectrical signals generated when a certain number of neuron cell populations repeatedly and periodically generate synchronous discharge carry a large amount of physiological activity information. The information-carrying bioelectric signal is an electroencephalogram signal. The electroencephalogram signal is a mixed wave with complex frequency components, the frequency range of the electroencephalogram signal is about 0.5-60 Hz generally, and the electroencephalogram signal is divided into five wave bands as shown in figure 2 according to the frequency difference in the medical field. The electroencephalogram signal multi-feature extraction based on wavelet packet transformation is adopted for electroencephalogram signals, and comprises the steps of determining signal frequency of the obtained electroencephalogram signals, calculating a decomposition scale j, obtaining wavelet packet sub-bands, reconstructing and obtaining 5 rhythm waves, calculating fast and slow wave energy and sample entropy, evaluating the wave energy and the sample entropy, and selecting feature vectors meeting requirements.

Extraction of electromyographic signal features, EMG is a time and space integration of the electrical activity of the epidermal muscles at the skin surface that can be collected by surface electrodes and avoids traumatic defects such as needle electrodes penetrating the muscles. Therefore, the bioelectric signal is the bioelectric signal of the neuromuscular system during the activity, which is guided and recorded from the surface of the muscle through the electrodes, and is mainly the combined effect of the electric activity of the superficial muscle and the nerve trunk. It has different degrees of correlation with the activity state and the function state of the muscle, so that the activity of the neuromuscular can be reflected to a certain degree. The myoelectric signal is a very weak signal, the amplitude of the myoelectric signal is l 00-5000 uV, the peak-peak value of the myoelectric signal is generally 0-6 mV, the root mean square is 0-l.5 mV, the frequency component of the generally useful signal is within the range of 0-500 Hz, and the main energy is concentrated within the range of 50-150 Hz.

For the raw sEMG measured via the surface electrodes, it is first subjected to 50Hz notching to eliminate power frequency interference. And then carrying out 10-500 Hz band-pass filtering on the signals through an IIR (infinite impulse response) system to complete the processing of the original signals. On the basis, the primary characteristic analysis of the sEMG is as follows:

in the aspect of time domain analysis, the method mainly comprises an integrated myoelectric value (iegm), a root mean square value (RMS), an absolute value integration, a zero crossing point number, a variance, a Willison amplitude, a time sequence model of the EMG signal, an EMG histogram and the like. Of these, the imeg and RMS values are most commonly used, and both can map the varying characteristics of sEMG signal amplitude in the time dimension, and the latter depends on the intrinsic relationship between muscle load factors and the physiological and biochemical processes of the muscle itself. In the aspect of frequency domain analysis, the main analysis method is to perform fast fourier transform on the sEMG signal to obtain a frequency spectrum or a power spectrum of the sEMG signal, which can reflect the change of the sEMG signal in different frequency components, so that the change of the sEMG signal can be better reflected in the frequency dimension. And obtaining a characteristic matrix of the electromyographic signal as an input matrix of a later model by extracting the characteristics of the electromyographic signal in two aspects of time domain and frequency domain.

And (3) extracting pulse signal characteristics, wherein in the process of a cardiac cycle, when a heart chamber contracts to eject blood, the blood is ejected into the aorta, and the blood is subjected to resistance of a vascular system to cause side pressure on the wall of the artery blood vessel to cause the aorta to expand. When ventricular diastolic ejection ceases, the aorta contracts. Such a contraction causes the pressure in the aorta to propagate in the form of a wave along with the blood flow in the arterial vessel wall, i.e., a pulse wave. The waveform change of the pulse wave can reflect the change conditions of various physiological and pathological information of human bodies, so the waveform characteristics of the pulse wave are important parameters for evaluating the state of the cardiovascular system. Fluctuation of the pulse wave is related to factors such as peripheral resistance of blood vessels, elasticity of the blood vessel walls, blood viscosity and the like, the blood vessel walls of healthy young people have good elasticity and small peripheral resistance, so that ascending branches and descending branches of the pulse wave are steep, and the main wave is represented as a high peak. Tidal waves are not apparent because of the lower velocity of reflected waves. The peak and trough of the dicrotic pulse are evident as the backflow of blood hits the aortic valve. Clinical studies have shown that pulse wave shape changes according to health and age, mainly reflected as changes in the height of tidal waves. The more stable characteristic point in the pulse wave waveform is the main wave, so the main wave is usually used as the characteristic point for studying pulse rate and pulse wave conduction time.

As a low-frequency non-stationary signal, the pulse wave signal needs to be preprocessed, and the burr and the baseline drift are removed mainly by utilizing wavelet transformation. The most remarkable characteristic of the non-stationary signal is the time-frequency local property, and the wavelet transformation can represent the local characteristics of the signal in a frequency domain and a time domain due to the characteristic of multi-resolution analysis, so that the abrupt change part and the noise of the signal can be effectively distinguished on different decomposition layers, and the noise elimination of the signal is realized. In calculating the pulse wave transit time based on the electrocardiography and the pulse wave, the main wave peak point of the pulse wave is often used as an end point. The differential double-threshold method has a good detection effect, so that the differential double-threshold method extracts the peak point of the pulse wave. The other pulse wave is characterized in that the change of the area of the pulse wave chart is used as the value of the pulse wave waveform characteristic quantity K. The K value is calculated as the area of one cycle of the pulse wave. The pulse wave waveform and area under different physiological and pathological states can be greatly changed, and the change can be represented by a K value. Fig. 4 shows pulse waveform diagrams in different physiological and pathological states. Wherein a corresponds to young healthy people, athletes or pregnant women, and has low vascular resistance and good artery elasticity, and K = 0.33 or so in clinical practice. b. c, d and e correspond to healthy middle-aged and young people, and are clinically manifested as moderate vascular resistance and arterial elasticity, wherein K = 0.34-0.39. f. g corresponds to middle-aged and elderly people, and the clinical manifestations are that vascular resistance is higher and artery elasticity is poor, and K =0.4~ 0.45. h corresponds to patients with severe hypertension and vascular atherosclerosis, and the clinical manifestations are that the vascular resistance is extremely high and the elasticity of arteries is extremely poor, and K =0.5 or so.

Blind source separation of cardiorespiratory signals

The heart and lung sounds are divided into heart sounds and lung sounds, i.e. sounds produced by the organs and organs of the human body during rhythm, and include information on physiological functions and pathological changes of the heart and blood vessels. The heart sounds also characterize part of the information about the systemic circulation and the cardiopulmonary circulation of the human body. The heart sound occurs in a specific period of regular beating of heart organs, and the tone and duration of the heart sound have certain rules. In general, a normal heart can produce four heart sounds, which we call: first, second, third and fourth heart sounds. Adults typically produce two types of heart sounds: a first heart sound and a second heart sound. Sometimes a third heart sound is produced in some children or young people, and a fourth heart sound may also occur in healthy people over the age of forty. In addition, abnormal heart sounds or noises may occur when the heart organs are diseased or an emergency occurs. The first heart sound occurs early in ventricular contraction, which marks the beginning of the ventricular systole. The second heart sound occurs at the end of ventricular systole, which marks the beginning of ventricular diastole. The lung sounds are the sounds generated by the respiratory system of a person during ventilation, and various diseases of the lungs of the person containing a great deal of information are reflected in the lung sounds of which the rule is not known yet. Respiratory sounds are generally divided into two categories: normal breathing sounds and additional sounds. Normal persons can hear three kinds of respiratory sounds, namely, a tubular respiratory sound (bronchial respiratory sound), an alveolar respiratory sound, and a bronchoalveolar respiratory sound. Clinically, abnormal breathing sounds are manifested as additive sounds, which are classified into continuous (aristocratic, wheezing and whistle) and intermittent (crackling and sputum-sounding).

In order to separate heart and lung sounds, an original signal model is firstly established, then short-time Fourier transform is carried out on the original signal to obtain an original signal spectrum, and the spectrum signal is decomposed into two sub non-negative matrixes W and H by using a non-negative matrix decomposition NMF technology. After the sub non-negative matrixes W and H are obtained, a given clustering algorithm is utilized to carry out classification training on each basis vector of the basis matrix W and the weight vector of the coefficient matrix H, and the basis vectors are divided into two types of signals: a heart sound spectrum signal and a lung sound spectrum signal. Meanwhile, the heart sound spectrum and lung sound spectrum signals on a time-frequency domain are converted into waveform signals on a time domain by utilizing an inverse short-time Fourier transform (ISTFT) technology. Finally, the two types of signals are used for the subsequent judgment and study. The specific separation process is shown in figure 5,

According to the vital sign multi-physiological signal fusion analysis method, the basic goal of information fusion is to deduce more information through information combination instead of any individual element appearing in input information, which is the result of optimal synergy, namely, the effectiveness of the whole sensing system is improved by utilizing the advantage of common or combined operation of a plurality of sensors. Generally, information fusion will take different forms depending on the different domains involved. According to different object processing levels, information fusion is often divided into: data level fusion, feature level fusion, and decision level fusion. Data level fusion is a low level fusion that retains as much information as possible and therefore has a high accuracy, but because of a large data processing amount, the processing time is long and the real-time performance is poor. Typically, the input to feature-level information fusion is the features acquired in the individual sensors, in which case less information is lost, which is also an advantage of feature-level information fusion. The decision-level fusion is a high-level fusion, before the fusion, a signal processing device of each sensor completes a decision or a classification task, and then a global optimal decision is made according to a certain criterion and the reliability of the decision, compared with the characteristic-level fusion, the decision-level fusion has the advantages that the loss amount of information is more, on the other hand, the decision-level fusion enables all input variables to have the same expression, and finally, the expandability of the information of all information sources can be better. Therefore, the research aims to adopt a decision-level fusion method to realize fusion analysis of signals such as electrocardio, electroencephalogram, myoelectricity, pulse, heart-lung sound and the like. In order to further improve the classification accuracy, the weighted voting method is used for decision-level fusion.

The SVM is established on the basis of the development of a calculation learning theory. Most linear classifiers use hyperplanes for classification. The SVM expands the concept of the hyperplane, and when the sample does not meet the linear separable condition, the SVM maps the feature space to a high-dimensional space to obtain a classification hyperplane. The SVM utilizes the classification interval to control the capacity of the classifier, so that the structural risk is minimum, and the generalization capability in a small sample space is enhanced. The SVM better solves the practical problems of small samples, nonlinearity, high dimensionality, local minimum points and the like, so the research selects the SVM as a base classifier, and provides the weighted SVM classifier shown in fig. 6 for identifying the physiological state of the sick and wounded. In the weighted SVM classifier in fig. 6, each physiological signal feature is first selected by using a correlation coefficient method, which specifically includes:

b. and selecting corresponding features with correlation coefficients higher than 0.1, inputting the features into each SVM classifier for training, respectively calculating the accuracy of the training set of each feature in the training process, and setting the accuracy of the training set with the training accuracy lower than 50% in the formula to zero in order to eliminate the influence of the features with poor classification performance on the classification result.

c. In the final classification experiment, the classification results were calculated. And finally, obtaining a final classification result by using a voting method, so that the final classification result is consistent with the class with the highest weight.

In order to support the signal acquisition, processing and wireless transmission described in this embodiment, this embodiment adopts a high-performance Coretex-M4 processor, a low-noise analog-to-digital converter suitable for biopotential measurement, a master-slave integrated wireless transmission module based on bluetooth 5.0, a power path selector, a lithium polymer battery charging management IC, and an anti-static surge protection device in hardware selection. In order to realize the functions described in the system structure, the hardware design structure of the equipment is as follows, the signal measurement unit realizes the measurement of human body signals and transmits the human body signals to the data acquisition unit, and in order to ensure the anti-interference performance and reliability of the system, the original signals acquired by the signal measurement unit firstly pass through a high-voltage resistant and anti-surge interface protection circuit to realize the filtering and protection of interference signals of high-voltage equipment such as a defibrillator and the like; the signal filtering unit simply filters some interference signals in the original signals collected by the signal measuring unit; the signals are processed by the interface protection circuit and the signal filter circuit and then respectively transmitted to the external high-precision special ADC and the ADC integrated in the CPU of the equipment for data conversion from analog quantity to data quantity of the signals. And the data is transmitted to a CPU of the equipment after ADC data conversion, and then the CPU of the equipment uploads the data to a host through controlling a low-power wireless transmission module for the host to analyze and process the data.

The power supply processing unit is internally provided with a power supply intelligent switching and automatic power supply management circuit, the circuit realizes the charging management of a device battery and the intelligent switching of a power supply in a charging state, when the electric quantity of the battery is insufficient, a power adapter is inserted, the power supply state of the device power supply is automatically switched to the power supply state of the adapter without influencing the normal use of the device, and when the adapter is pulled out after the battery is fully charged, the device can be automatically switched to the power supply of the battery; when the equipment is in a standby state, the power management automatically enters a low-power-consumption management mode, the equipment is in a low-power-consumption power supply state, and when the equipment is triggered to be in a normal working mode, the power management enters the normal working mode, so that the intelligent management of the equipment power is realized.

Although the present invention has been described with reference to a preferred embodiment, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. The method for extracting and fusing the multiple physiological signals of the vital signs is characterized by comprising the following steps of:

2) preprocessing each signal and extracting characteristics in the step 1);

c. and obtaining a final classification result by using a voting method.

2. The method for extracting and fusing multi-physiological signals of vital signs according to claim 1, wherein: the preprocessing and feature extraction of the electrocardiosignals comprise electrocardiosignal denoising preprocessing, high-quality electrocardiosignal screening and electrocardiosignal feature extraction, wherein the electrocardiosignal denoising preprocessing adopts a soft threshold value method to process low-scale wavelet coefficients d1 and d2, and high-frequency noise is eliminated through a large-amplitude attenuation coefficient method; and filtering the scale 3 wavelet coefficient containing the important input signal by adopting a soft and hard threshold compromise algorithm to obtain a clean electrocardiosignal.

3. The method for multi-physiological signal extraction and fusion analysis of vital signs according to claim 1, wherein: extracting the characteristics of the electrocardiosignals, and finishing the detection of singular points of the ECG signals based on a biorthogonal quadratic B-spline wavelet transform method; improving the positioning precision of the R wave by adopting a dynamic valve value method, analyzing a wavelet coefficient D4, extracting an extreme value pair, obtaining the extreme value pair for accurately positioning the R wave by adopting threshold processing and filtering, restoring the position of the extreme value pair to a reconstructed ECG signal, finding the maximum value in the maximum value interval, and obtaining the R wave positioning; after the R wave is positioned, Q, S waves can be positioned by searching a minimum value point near the R wave according to the position of the R wave;

the length of a detection interval is controlled by taking the positions of two adjacent R waves as a marker post and the mean value of an RR interval, so that two heart beats between the two adjacent R waves, respective T waves and P waves are accurately detected; and measuring respective detection areas of the T wave and the P wave to complete accurate positioning of P, Q, R, S, T waves of the ECG signal.

4. The method for multi-physiological signal extraction and fusion analysis of vital signs according to claim 1, wherein: extracting the basic features and the characteristics of the electroencephalogram signals, namely extracting the multiple features of the electroencephalogram signals based on wavelet packet transformation; determining signal frequency of the obtained electroencephalogram signal, calculating a decomposition scale j, obtaining a wavelet packet sub-band, reconstructing and obtaining 5 rhythm waves, calculating fast and slow wave energy and sample entropy, and evaluating the wave energy and the sample entropy to select a feature vector meeting requirements.

5. The method for multi-physiological signal extraction and fusion analysis of vital signs according to claim 1, wherein: extracting an iEMG value and an RMS value from the electromyographic signals to be used as electromyographic time domain characteristics; and carrying out fast Fourier transform on the sEMG signal to obtain the electromyographic frequency domain characteristics.

6. The method for multi-physiological signal extraction and fusion analysis of vital signs according to claim 1, wherein the pulse signal is processed by wavelet transform to remove spike and baseline wander, and differential dual threshold method is used to extract peak points of the pulse wave; and taking the change of the area of the pulse wave chart as the value of the pulse wave waveform characteristic quantity K.

7. The method for multi-physiological signal extraction and fusion analysis of vital signs according to claim 1, wherein: the extraction of the heart-lung sound signals comprises the following steps;

c) and (3) carrying out classification training on each basis vector of the basis matrix W and the weight vector of the coefficient matrix H by using a clustering algorithm, and dividing the weight vectors into two types of signals, namely heart sound spectrum signals and lung sound spectrum signals:

8. The wireless vital sign monitoring device using the method for extracting and fusing multiple physiological signals for vital sign analysis according to any one of claims 1 to 7, wherein: the device comprises a signal measuring unit, a data processing unit, a data transmission unit and a power management unit; the signal acquisition unit comprises electroencephalogram, electrocardio and myoelectricity signal acquisition through an electrode, blood pressure acquisition through a cuff with a high-precision pressure sensor, respiration measurement through a chest elastic band and body temperature acquisition through a contact type body temperature measurement sensor.

9. The wireless vital sign monitoring device of claim 8, wherein:

the data processing unit preprocesses the acquired electroencephalogram, electrocardio, heart sound and electromyogram signals through a signal filtering unit and an external ADC signal conversion unit, wherein the signal filtering unit simply filters some interference signals in the original signals acquired by the signal measuring unit; the signal is processed by the interface protection circuit and the signal filter circuit and then is respectively transmitted to the external high-precision special ADC and the ADC integrated in the CPU of the device to perform data conversion from the analog quantity to the data quantity of the signal, the data is transmitted to the CPU of the device after the data conversion of the ADC, and then the CPU of the device uploads the data to the host computer by controlling the low-power wireless transmission module for the host computer to perform data analysis and processing.