WO2022062237A1 - Smart wearable fetal movement monitoring system - Google Patents

Abstract

A smart wearable fetal movement monitoring system, comprising: a data collection module, a smart terminal, a data storage and analysis and determination module deployed on a cloud server, and a data management module. The system deploys the data storage and analysis and determination module on the cloud server. Compared with the conventional approach of embedding such a module into a hardware terminal or front-end software, the approach of the invention has significant advantages, including good security and confidentiality, convenient iterative updating of algorithms and expansion of computing performance, and good support for third-party libraries such as math or AI algorithm libraries. In addition to monitoring fetal movement count, the system can also monitor a number of fetal movement evaluation indicators, including the manner of fetal movement (such as kicking, continuous movement, intense movement, etc.), the longest duration of continuous fetal movement, the proportion of intense fetal movement within said longest duration, the longest still period of the fetus, and fetal activity level. The invention, by means of such indicators, quantitatively describes for the first time the perceptions of a pregnant women of the intensity, characteristics, and duration of fetal movement, and thus has important medical significance as reference data.

Description

Smart wearable fetal movement monitoring system technical field

The invention relates to the technical field of fetal movement detection, in particular to an intelligent wearable fetal movement monitoring system.

Background technique

Fetal movement is the most intuitive feeling that pregnant women have about the health of the fetus in the uterus. When the fetus suffers from intrauterine injury, growth retardation or even death, pregnant women can detect abnormal changes in fetal movement. Most of the guidelines involving fetal movement self-monitoring behavior affirm the importance of fetal movement self-monitoring. In China, the "Guidelines for Pre-pregnancy and Pregnancy Health Care (2018)" recommend that pregnant women start counting fetal movements at 29-30 weeks ^of pregnancy1, and most foreign guidelines recommend pregnant women at 28 weeks. After that, monitor fetal movement every day ² . Decreased fetal movement is often the first sign of fetal ^death3 . The 2018 Australian/New Zealand clinical practice guidelines point out that pregnant women should not only pay attention to the number of fetal movements, but also pay more attention to the intensity, characteristics and duration of fetal movement4 ^. Generally speaking, the longest duration of fetal movement is more than 5 minutes, and the proportion of strong fetal movement during the longest duration exceeds 80%, which means that the fetus has been in an active state during this time period, and the intensity of the movement is too high. When the fetal movement is abnormally active, strong and lasts too long, pregnant women and medical staff should be cautious to prevent the occurrence of umbilical cord around the neck and insufficient oxygen supply to the fetus. A New Zealand case-control study of 155 stillbirths and 310 live births showed that sudden, intense fetal movements over a short period of time were associated with a nearly seven-fold increased risk of stillbirth. Sudden, frequent and intense fetal movement, especially the subsequent reduction or disappearance of fetal movement, is a manifestation of acute fetal ^distress . The longest static time of the fetus is more than 50min, and the fetal activity is less than 10%, which means that the number of fetal movements is less. If such a situation occurs frequently, pregnant women and medical staff should pay great attention to it. Frequently reduced fetal movement is closely related to preterm birth or even stillbirth. Most stillbirths are accompanied by frequent reduced fetal movement within 3-4 days, and 55% of pregnant women who experience stillbirth can noticeably perceive frequent reduced ^fetal movement7 . Decreased fetal movement is thought to be associated with adverse outcomes such as fetal growth restriction, stillbirth, and a ⁴ -fold increased risk of stillbirth in those with reduced fetal movement. Preterm birth and other adverse pregnancy outcomes are closely related. There is an increased likelihood of adverse pregnancy outcomes such as perinatal brain injury and neonatal hypoxic ischemic encephalopathy9 ^;10 . Taking into account the number of fetal movements, the longest fetal static time, and the indicators of fetal activity, when frequent fetal movements occur less frequently, regardless of the results of ultrasound assessment, it means that pregnant women have a high risk of ^placental dysfunction7, and further medical diagnosis is required.

Therefore, in addition to the number of fetal movements, the monitoring of more forms of fetal movement indicators (such as the longest fetal static time, fetal activity, etc.) has important medical reference significance. level, it is difficult to fully reflect the status of the fetus.

Citation:

[1] Obstetrics Group, Obstetrics and Gynecology Branch of Chinese Medical Association. Guidelines for health care before and during pregnancy (2018) [J]. Chinese Journal of Obstetrics and Gynecology, 2018, 53(1): 7-13.

[2] Zhang Wen, Zhang Jing. Research status and guideline recommendations of maternal fetal movement self-monitoring behavior [J]. China Maternal and Child Health Research, 2017, 28(2): 213-215.

[3]Jf F.A kick from within-fetal movement counting and the canceled progress in antenatal care[J].J Perinat Med,2004,32(1):13-24.

[4]Daly L M, Gardener G, Bowring V, et al.Care of pregnant women with decreased fetal movements:Update of a clinical practice guideline for Australia and New Zealand[J].Australian and New Zealand Journal of Obstetrics and Gynaecology, 2018, 58(4): 463-468.

[5] Ryo E, Nishihara K, Matsumoto S, et al.A new method for long-term home monitoring of fetal movement by pregnant women themselves[J].Med Eng Phys,2012,34(5):566-72.

[6] T S, Jmd T, Ea M, et al. Maternal perception of fetal activity and late stillbirth risk: findings from the auckland stillbirth study[J].Birth,2011,38(4):311-316.

[7]Scala C, Bhide A, Familiari A, et al.Number of episodes of reduced fetal movement at term:association with adverse perinatal outcome[J].Am J Obstet Gynecol,2015,213(5):678e1-6.

[8] Wang Menglu, Chen Qian. The etiology and prevention of stillbirth [J]. Chinese Journal of Practical Gynecology and Obstetrics, 2017, 33(11): 1121-1125.

[9] M T, H F, S S, et al. Brain damage caused by severe fetomaternal hemorrhage [J]. Pediatrics International, 2010, 52(2): 301-304.

[10] Hei Mingyan. Fetal movement reduction in late pregnancy and neonatal hypoxic-ischemic brain injury [J]. Chinese Journal of Practical Pediatrics, 2015, 30(4): 311-316.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide an intelligent wearable fetal movement monitoring system in view of the above-mentioned deficiencies in the prior art.

In order to solve the above technical problems, the technical solution adopted in the present invention is: an intelligent wearable fetal movement monitoring system, comprising: a data acquisition module, an intelligent terminal, a data storage and analysis and determination module and a data management module deployed on a cloud server;

The data acquisition module includes N acquisition units arranged on the abdominal wall of the pregnant woman at intervals for collecting vibration signals of the abdominal wall of the pregnant woman, so as to perform parallel acquisition of N channel data;

The intelligent terminal is used for transmitting the multi-channel data collected by the data collection module to the data storage and analysis and determination module;

The data storage and analysis and determination module includes a data storage unit and an analysis and determination unit, the data storage unit receives and stores the multi-channel data sent by the intelligent terminal, and the analysis and determination unit analyzes and determines the multi-channel data to The vibration signal distinguishes the fetal movement signal and the noise signal, and further distinguishes and determines the specific form of the fetal movement signal and the noise signal to form a classification result; wherein, the specific form of the fetal movement signal includes at least the percussion form of fetal movement and the strong form of fetal movement, and the noise signal. The specific forms of noise include at least breathing noise, sneezing noise and body motion noise;

The data management module includes a statistical unit and an application terminal, and the data management module receives the classification result sent by the data storage and analysis and determination module, and accordingly forms a multi-dimensional fetal movement index per unit monitoring time;

The application terminal displays the multi-dimensional fetal movement index obtained by the statistical unit.

Preferably, the collection unit is a wearable pressure sensor disposed on the abdominal wall of the pregnant woman.

Preferably, the application terminal includes a management background and a personal user terminal.

Preferably, the multi-dimensional fetal movement index includes at least the number of fetal movements per unit monitoring time, the form of fetal movement, the longest duration of fetal movement, the proportion of strong fetal movement in the longest duration, the longest fetal static time, and fetal activity;

Wherein, the longest duration of the fetal movement is: the total time occupied by the interval with the longest duration among the determination unit intervals that are continuously determined as fetal movement signals within the unit monitoring time;

The proportion of strong fetal movement within the longest duration is: within the unit monitoring time, in the longest duration interval of fetal movement, the percentage of the interval determined to be a strong form of fetal movement;

The longest static time of the fetus is: the total time occupied by the interval with the longest duration among the determination unit intervals that are continuously determined to be noise signals within the unit monitoring time;

The fetal activity is: the percentage of the determination units determined as fetal movement signals in all determination units in the unit monitoring time in the unit monitoring time.

Preferably, the method for performing fetal movement monitoring by the smart wearable fetal movement monitoring system comprises the following steps:

S1, carry out the parallel collection of the vibration signal data of N channels to the abdominal wall of pregnant women by the data collection module;

S2. The intelligent terminal wirelessly transmits the multi-pass data collected by the data collection module to the data storage and analysis and determination module;

S3. The data storage unit in the data storage and analysis and determination module receives and stores the multi-channel data sent by the intelligent terminal, and then the analysis and determination unit in the data storage and analysis and determination module analyzes and determines the multi-channel data, The vibration signal is distinguished from the fetal movement signal and the noise signal, and the specific form of the fetal movement signal and the noise signal is further distinguished and determined to form a classification result;

S4, the data management module receives the classification result sent by the data storage and analysis and determination module, and accordingly forms a multi-dimensional fetal movement index per unit monitoring time;

S5. The management background and the personal user terminal in the application terminal display the obtained multi-dimensional fetal movement index for the user to obtain.

Preferably, the method for analyzing and determining the multi-channel data by the analyzing and determining unit in the step S3 includes the following steps:

1) Data preprocessing;

2) For the preprocessed data, firstly, the data of the N channels received in the unit monitoring time is cut and segmented according to the timeline to form multiple independent judgment units, and then the multiple judgment units are divided one by one according to the timeline. Analyze and determine through the following steps;

3) by multi-channel feature analysis, the vibration signal is firstly judged by distinguishing the fetal movement signal and the noise signal, and then the specific form of the fetal movement signal and the noise signal is discriminated and judged to form a classification result, and then the analysis and judgement of the next judgment unit is carried out;

4) Repeat the above step 3) until the analysis and determination of all the determination units within the unit monitoring time are completed.

Preferably, the step 1) specifically includes: first, the vibration signal is smoothed by a method including at least wavelet threshold denoising and Butterworth filtering, and then a baseline drift correction is performed by using an asymmetric least squares baseline correction method .

Preferably, the step 3) specifically includes:

3-1) Perform multi-channel feature extraction on a single decision unit, including:

3-1-1) Extract the Pearson correlation coefficient between the N channels, then calculate the mean value of the Pearson correlation coefficient for each channel for other channels, and denote it as the Pearson correlation coefficient mean value ρ _n ;

3-1-2) Calculate the peak-to-peak quotient of each channel in the N channels: According to the sampling frequency, set the minimum horizontal unit distance between peak-to-peak values, and at this distance, calculate the maximum peak value and minimum peak value of each channel The ratio, denoted as peakDivision _N ;

3-1-3) Perform spectrum analysis and normalization on the N channel signals, and calculate the average value amp _n of the corresponding amplitudes of the N channel signals at different frequencies, and count the number of channels whose amp _n is greater than the amplitude threshold thresholdAmp, denoted as count _{(amp>thresholdAmp)} ;

3-1-4) Matrix feature analysis: In a single judgment unit, the vibration signal data of N channel signals are formed into an M*N matrix, where M represents the product of the sampling frequency f and the judgment unit duration t, that is, M=f*t ; For this matrix, meanRowValue represents the mean value of the row vector, and index represents the index of the row vector, taking the following characteristics:

a) The difference between the maximum value and the minimum value of the row vector mean

r _max-min , r _max-min = meanRowValue _max -meanRowValue _min ,

Among them, meanRowValue _max represents the maximum value of the mean value of the row vector, and meanRowValue _min represents the minimum value of the mean value of the row vector;

[Corrected 01.02.2021 according to Rule 26]
b) The index difference between the index of the maximum value of the row vector mean and the index of the minimum value of the row vector mean index _max-min ,

Among them, index _max represents the index of the maximum value of the row vector mean value,

The index of the minimum value of the mean value of the row vector;

2-2) carry out the concrete classification of fetal movement signal and noise signal to the vibration signal of this determination unit according to the multi-channel feature obtained in step 2-1), including:

3-2-1) Set two Pearson correlation coefficient judgment thresholds A and B, where 0<A<B, and compare the minimum value ρ _min in the mean value ρ _n of the Pearson correlation coefficient of all channels with A and B. Compare, and then judge according to the comparison results according to the following different steps;

According to statistical experience, the Pearson correlation coefficient range is divided as follows: 0.8-1.0, very strong correlation; 0.6-0.8, strong correlation; 0.4-0.6, moderate correlation; 0.2-0.4, weak correlation; 0.0-0.2, very strong correlation Weak correlation or irrelevance; <0, negative correlation; two thresholds A and B need to be determined and improved by comprehensively considering the number of channels N and the above statistical experience;

3-2-2) When ρ _min >B, count the peak-to-peak quotient peakDivisionN of each channel in the N channels;

1. If the peak-to-peak quotient of at least one channel is greater than the set threshold value P _T , then it is determined as the fetal movement in the form of percussion, and the classification result is output, otherwise the following steps continue to be determined;

According to the experimental results, the threshold P _T needs to be determined and perfected according to the sensing characteristics (sensitivity, linearity, etc.) of the sensing device; generally speaking, 2.0≤P _T ≤2.5;

II. Count the number of count (amp>thresholdAmp ₎ , if count _{(amp>thresholdAmp)} = J, it is determined to be breathing noise, and the classification result is output, where J is a positive integer not greater than N; otherwise, proceed according to the following steps determination;

III. Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is determined as sneezing noise, otherwise it is determined as body motion noise; output the classification result, where I is the proportional coefficient , 1≤I≤2.5;

According to the experimental results, the threshold value thresholdValue needs to be determined and perfected according to the sensing characteristics of the sensing device (response curve, linearity, etc.), and the proportional coefficient I needs to be determined and perfected according to the performance of the sample. Generally speaking, 1≤I≤2.5;

3-2-3) When ρ _min <A, if the number of Pearson correlation coefficients less than A is not greater than k in the Pearson correlation coefficients of all channels, it is determined as percussion fetal movement, otherwise it is determined as strong fetal movement , output the classification result; where k>N/2; according to the experimental results, the value of k needs to be determined and perfected according to the number of channels N and the performance of the sample, in general, k>N/2;

3-2-4) When A < ρ _min < B, then count the peak-to-peak quotient peakDivisionN of each channel in the N channels, if there is at least one channel whose peak-to-peak quotient is greater than the set threshold P _T , then count again. Determine according to the following step a), otherwise according to the following step b) to determine;

a) If among the Pearson correlation coefficients of all channels, the number of Pearson correlation coefficients less than A is not greater than k, it is determined as a percussion form of fetal movement, otherwise it is determined as a strong form of fetal movement, and the classification result is output; where k>N/ 2;

b) Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is judged as sneezing noise, otherwise it is judged as body motion noise; output the classification result, where I is the proportional coefficient , 1≤I≤2.5.

Preferably, the acquisition units include four uniformly spaced laterally arranged on the abdominal wall of the pregnant woman, so as to perform parallel acquisition of 4-channel data.

Preferably, the step 3) specifically includes:

3-1) Perform multi-channel feature extraction on a single decision unit, including:

3-1-1) Extract the Pearson correlation coefficient between the 4 channels. The calculation formula of the Pearson correlation coefficient is:

ρ _X,Y =cov(X,Y)/σ _X σ _Y =E((X-μ _X )(Y-μ _Y ))/σ _X σ _Y ;

Among them, cov(X, Y) represents the covariance between two variables X, Y, σ _X , σ _Y represents the standard deviation of the two variables X, Y; s0, s1, s2, s3 represent the four-channel passing through For the preprocessed vibration signal data, there are 6 Pearson correlation coefficients between the four channels, namely ρ _0,1 , ρ _0,2 , ρ _0,3 , ρ _1,2 , ρ _1,3 , ρ _{2 , 3.} Calculate the mean of the Pearson correlation coefficient of each channel for other channels, namely:

ρ ₀ =1/3(ρ _0,1 +ρ _0,2 +ρ _0,3 );

ρ ₁ =1/3(ρ _0,1 +ρ _1,2 +ρ _1,3 );

ρ ₂ =1/3(ρ _0,2 +ρ _1,2 +ρ _2,3 );

ρ ₃ =1/3(ρ _0,3 +ρ _1,3 +ρ _2,3 );

3-1-2) Calculate the peak-to-peak quotient of each of the 4 channels: According to the sampling frequency, set the minimum horizontal unit distance between the peak and peak values, and at this distance, calculate the maximum peak value and the minimum peak value in each channel The ratio, denoted as peakDivision ₀ , peakDivision ₁ , peakDivision ₂ , peakDivision ₃ ;

3-1-3) Perform spectrum analysis and normalization on the four-channel signal, and calculate the average value amp _n of the corresponding amplitudes of the four channel signals at different frequencies, and count the number of channels whose amp _n is greater than the amplitude threshold thresholdAmp, denoted as count _{( amp>thresholdAmp)} ;

3-1-4) Matrix feature analysis: In a single determination unit, the vibration signals of the 4 channel signals are formed into an M*4 matrix, where M represents the product of the sampling frequency f and the determination unit duration t, that is, M=f*t; In this matrix, meanRowValue represents the mean value of the row vector, and index represents the index of the row vector, and takes the following characteristics:

a) The difference between the maximum value and the minimum value of the row vector mean value r _max-min ,

r _max-min = meanRowValue _max -meanRowValue _min ,

Among them, meanRowValue _max represents the maximum value of the mean value of the row vector, and meanRowValue _min represents the minimum value of the mean value of the row vector;

[Corrected 01.02.2021 according to Rule 26]
b) The index difference between the index of the maximum value of the row vector mean and the index of the minimum value of the row vector mean index _max-min ,

Among them, index _max represents the index of the maximum value of the row vector mean value,

The index of the minimum value of the mean value of the row vector;

3-2) carry out the concrete classification of fetal movement signal and noise signal to the vibration signal of this judgment unit according to the multi-channel feature obtained in step 3-1), including:

3-2-1) Set two Pearson correlation coefficient judgment thresholds A and B, where A=0.4, B=0.6; and compare the minimum value ρ _min in the mean value ρ _n of the Pearson correlation coefficient of all channels with A Compare with B, and then judge according to the following different steps according to the comparison result;

3-2-2) (ρ ₀ , ρ ₁ , ρ ₂ , ρ ₃ ) when _min > 0.6, count the peaks of each of the 4 channels

peak quotient peakDivisionN;

1. If there is a peak-to _- peak quotient of at least one channel greater than the set threshold PT, _PT = 2, then it is judged that the fetal movement in the form of percussion, the output classification result, otherwise continue to judge according to the following steps;

II. Count the number of count (amp>thresholdAmp ₎ , if count _{(amp>thresholdAmp)} =J, J=2, then it is judged as breathing noise, and the classification result is output; otherwise, continue to judge according to the following steps;

III. Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is determined as sneezing noise, otherwise it is determined as body motion noise; output the classification result, where I=1.5;

3-2-3) When ρ _min <A, if the number of Pearson correlation coefficients less than A is not greater than k in the Pearson correlation coefficients of all channels, it is determined as percussion fetal movement, otherwise it is determined as strong fetal movement , output the classification result; where k=3;

3-2-4) When A < ρ _min < B, then count the peak-to-peak quotient peakDivisionN of each channel in the N channels. If there is at least one channel whose peak-to-peak quotient is greater than the set threshold 2, press again. The following steps a) are judged, otherwise the following steps b) are judged;

a) If among the Pearson correlation coefficients of all channels, the number of Pearson correlation coefficients less than A is not greater than k, it is determined as a percussion form of fetal movement, otherwise it is determined as a strong form of fetal movement, and the classification result is output; wherein k=3;

b) Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is judged as sneezing noise, otherwise it is judged as body motion noise; output the classification result, where I=1.5.

The beneficial effects of the present invention are:

1. The present invention deploys the data storage and analysis and determination module on the cloud server. Compared with the traditional method of embedding it in the hardware terminal or front-end software, the significant advantages of doing so are: good security and confidentiality, convenient iterative update of the algorithm, and easy calculation. Convenient performance expansion, good support for third-party libraries such as math library/AI algorithm library;

2. The present invention adopts the design method of multi-channel pressure sensor to collect the abdominal wall pressure signal of pregnant women, the number of channels is easy to expand, the extracted pressure signal is perfect and comprehensive, it is easy to distinguish and identify various signal features, and the accuracy and reliability are good;

3. The present invention can realize the obvious distinction and classification of different fetal movement forms such as tapping, strong, continuous and other noises, as well as noises such as body movement, sneezing, laughter, etc., through the fusion and comparison analysis algorithm based on the characteristics of multi-channel pressure signals. In the process of fetal movement monitoring, the determination of fetal movement forms and the identification of noise are more comprehensive and complete;

4. In addition to the number of fetal movements, the present invention can also include fetal movement forms (such as percussion, continuous, strong, etc.), the longest duration of fetal movement, the proportion of strong fetal movement in the longest duration, the longest fetal static time, and fetal activity. For the first time, these indicators quantitatively describe pregnant women's perception of fetal movement intensity, characteristics, and duration, and have important medical reference significance.

Description of drawings

Fig. 1 is the principle block diagram of the intelligent wearable fetal movement monitoring system of the present invention;

Fig. 2 is the schematic diagram of the 4-channel pressure sensor signal acquisition of the present invention;

Fig. 3 is the working flow chart of the intelligent wearable fetal movement monitoring system of the present invention;

4 is a preprocessing effect diagram of a breathing signal in a resting state in an embodiment of the present invention;

Fig. 5 is the signal preprocessing effect diagram under the fetal movement trigger state in the embodiment of the present invention;

6 is a waveform diagram of several representative signals enumerated in the present invention;

7 is a matrix diagram formed by pressure data of four-channel signals in an embodiment of the present invention;

FIG. 8 is a flowchart of specific classification of fetal movement signals and noise signals in an embodiment of the present invention.

detailed description

The present invention will be further described in detail below with reference to the embodiments, so that those skilled in the art can implement according to the description.

It should be understood that terms such as "having", "comprising" and "including" as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

Referring to FIG. 1 , an intelligent wearable fetal movement monitoring system in this embodiment includes: a data acquisition module, an intelligent terminal, a data storage and analysis and determination module and a data management module deployed on a cloud server.

1. Data acquisition module

The data collection module includes N collection units arranged on the abdominal wall of the pregnant woman at intervals and used for collecting vibration signals of the abdominal wall of the pregnant woman, so as to perform parallel data collection of N channels.

In this embodiment, the acquisition unit is a wearable pressure sensor disposed on the abdominal wall of pregnant women. In this embodiment, parallel multi-channel pressure sensors are used to collect the pressure distribution of the abdominal wall of pregnant women, which is convenient for expansion and maintenance. The number of sensors can be flexibly configured according to the sensing characteristics of the device (such as load area, sensitivity, etc.), as long as it can reflect the pressure trend changes in the entire abdominal wall range. If the number of sensors is too small, there will be a lack of signal capture. Too many sensors will lead to redundant signal acquisition, large data volume, and increased cost. In a preferred embodiment, the number of sensors should be at least not less than 3, which can cover the left, middle and right regions of the abdominal wall of the pregnant woman. In the following embodiments, a 4-channel pressure sensor is used for description (as shown in Figure 2). S0, S1, S2, and S3 respectively represent 4-channel pressure sensors, which are evenly distributed laterally on the abdominal wall of pregnant women. This distribution method can minimize fetal movement. The packaging area of the monitoring belt is small, and the skin area in direct contact with pregnant women is small, which can significantly improve the wearing comfort. In this distribution mode, the fetal movement signal collected at the far position in the longitudinal direction will be slightly weakened, and this problem can be well solved by means of data preprocessing in the present invention. When there is no fetal movement, S0, S1, S2, and S3 collect the normal abdominal wall breathing signals of pregnant women; when the fetal movement is in the form of "knocking", the triggered fetal movement is located at f0, and the fetal movement wave signal can be detected by S0 , S1 sensor acquisition, the signal strength received by S2, S3 is very weak; when the fetal movement form is "strong", it can trigger multiple areas of the abdominal wall of pregnant women, such as f1_a and f1_b areas, the fetal movement signal can be collected by S0, S3 sensors , the signal strength received by S1 and S2 is very weak.

In a preferred embodiment, the transmission between the sensor and the main control circuit can use a flexible cable or a fabric wire, which has the characteristics of invisibility, flattening, and flexibility. The sensor can be integrated into existing products such as abdominal support belt, fetal monitoring belt or elastic fabric, or the circuit can be encapsulated in silicone material, which can significantly improve the comfort and breathability of pregnant women when they wear it for a long time.

The acquisition frequency of the sensor depends on the situation. Generally speaking, it is between 10Hz and 100Hz. If it is too low, it is not conducive to the extraction and analysis of signal characteristics. If it is too high, it will put forward greater requirements on data transmission stability, battery power supply, and server storage space.

2. Smart terminal

The intelligent terminal is used for transmitting the multi-channel data collected by the data collection module to the data storage and analysis and determination module. The smart terminal is selected as a smart phone or tablet, etc., and the smart terminal is connected to the sensor, data storage and analysis and determination module by wireless communication (Bluetooth or WIFI, etc.).

3. Data storage and analysis and judgment module

The data storage and analysis and determination module includes a data storage unit and an analysis and determination unit, the data storage unit receives and stores the multi-channel data sent by the intelligent terminal, and the analysis and determination unit analyzes and determines the multi-channel data to The vibration signal distinguishes the fetal movement signal and the noise signal, and further distinguishes and determines the specific form of the fetal movement signal and the noise signal to form a classification result; wherein, the specific form of the fetal movement signal includes at least the percussion form of fetal movement and the strong form of fetal movement, and the noise signal. The specific forms of noise include at least breathing noise, sneezing noise and body motion noise.

The data storage and analysis and determination module is deployed on the cloud server, and the intelligent terminal uses bluetooth or wifi technology to upload to the cloud server and store it; The significant advantages of the traditional method in front-end software are: good security and confidentiality, convenient iterative update of algorithms, convenient expansion of computing performance, and good support for third-party libraries such as math library/AI algorithm library.

In general, fetal movement needs to be monitored 3 times a day for 1 hour each time. With the increase of end users and the increase of unit monitoring time, the data volume will require higher and higher storage space. The invention stores the data in the cloud server, which not only facilitates the elastic expansion of the storage capacity, but also facilitates the security and privacy protection of the user data. These factors are of great significance for establishing a fetal movement database. At present, there is no authoritative and comprehensive fetal movement database in the field of maternal and fetal health. .

4. Data management module

The data management module includes a statistical unit and an application terminal, and the classification result of the data storage and analysis and determination module is sent back to the data management module, and the data management module receives the classification result sent by the data storage and analysis and determination module, and according to This forms a multi-dimensional fetal movement indicator within the unit monitoring time.

The application terminal displays the multi-dimensional fetal movement indicators obtained by the statistical unit. The application terminal may include a management background and a personal user terminal (such as a mobile phone or a computer, etc.) The user terminal can be used by a single pregnant woman to know her fetal movement assessment.

The multi-dimensional fetal movement index includes at least the number of fetal movements within the unit monitoring time, the form of fetal movement, the longest duration of fetal movement, the proportion of strong fetal movement in the longest duration, the longest fetal static time and fetal activity, specifically:

The longest duration of fetal movement is: the total time occupied by the interval with the longest duration among the determination unit intervals continuously determined as fetal movement signals within a unit monitoring time. For example, the total length of the measurement data is 1 hour, and the single judgment unit is 6s. In this 1 hour, up to 10 judgment units are continuously judged as "true", and the longest duration of fetal movement is (6s*10=)60s.

The proportion of strong fetal movement within the longest duration is: the percentage of the interval determined to be a strong form of fetal movement in the longest duration interval of fetal movement within the unit monitoring time. For example, the longest duration of fetal movement is 60s. Among the 10 consecutive “true” determination units, there are 3 determination units in which the fetal movement form is “strong”, and the proportion of strong fetal movement in the longest duration is 3/ 10=30%.

The longest static time of the fetus is: the total time occupied by the interval with the longest duration among the determination unit intervals that are continuously determined to be noise signals within the unit monitoring time. For example: the total length of the measurement data is 1 hour, and the single judgment unit is 6s. In this 1 hour, at most 100 judgment units are continuously judged as "false", that is, no fetal movement signal is detected, then the longest fetal static time is (6s*100 =) 10min.

The fetal activity is: the percentage of the determination units determined as fetal movement signals in all determination units in the unit monitoring time in the unit monitoring time. For example, the total length of the measurement data is 1 hour, and there are 600 units to be judged. In this 1 hour, a total of 150 judgment units are judged as "true", and the fetal activity is 150/600=25%.

The above fetal movement indicators can effectively characterize the subjective feelings of pregnant women on the intensity, characteristics and duration of fetal movement. The occurrence of too low and other conditions urgently need the attention of doctors or pregnant women.

In addition to the number of fetal movements, the present invention can also include the fetal movement form (such as percussion, continuous, strong, etc.), the longest duration of fetal movement, the proportion of strong fetal movement in the longest duration, the longest fetal static time, and fetal activity. These indicators quantitatively describe for the first time pregnant women's perception of fetal movement intensity, characteristics, and duration, and have important medical reference significance.

In an embodiment, referring to FIG. 3 , the method for performing fetal movement monitoring by the smart wearable fetal movement monitoring system includes the following steps:

S1, carry out the parallel collection of the vibration signal data of N channels to the abdominal wall of pregnant women by the data collection module;

S2. The intelligent terminal wirelessly transmits the multi-pass data collected by the data collection module to the data storage and analysis and determination module;

S3. The data storage unit in the data storage and analysis and determination module receives and stores the multi-channel data sent by the intelligent terminal, and then the analysis and determination unit in the data storage and analysis and determination module analyzes and determines the multi-channel data, The vibration signal is distinguished from the fetal movement signal and the noise signal, and the specific form of the fetal movement signal and the noise signal is further distinguished and determined to form a classification result;

S4, the data management module receives the classification result sent by the data storage and analysis and determination module, and accordingly forms a multi-dimensional fetal movement index per unit monitoring time;

S5. The management background and the personal user terminal in the application terminal display the obtained multi-dimensional fetal movement index for the user to obtain.

Further, in the described step S3, the method that the analysis and determination unit analyzes and determines the multi-channel data comprises the following steps:

1) Data preprocessing;

2) For the preprocessed data, firstly, the data of the N channels received in the unit monitoring time is cut and segmented according to the timeline to form multiple independent judgment units, and then the multiple judgment units are divided one by one according to the timeline. Analyze and determine through the following steps;

3) Through multi-channel feature analysis, the vibration signal is firstly judged to distinguish the fetal movement signal and the noise signal, and then the specific form of the fetal movement signal and the noise signal is discriminated and judged to form a classification result, and then the analysis and judgment of the next judgment unit is carried out;

4) Repeat the above step 3) until the analysis and determination of all the determination units within the unit monitoring time are completed.

In this embodiment, step 1) specifically includes: first, the vibration signal is smoothed by a method including at least wavelet threshold denoising and Butterworth filtering, and then the baseline drift correction is performed by using the asymmetric least squares baseline correction method .

The hardware acquisition circuit has power frequency interference and high-frequency noise. Due to its own characteristics and changes in the external environment, the data collected by the pressure sensor has a baseline drift. The noise existing in these hardware modules is unavoidable. Such factors are important for subsequent feature extraction and model design. Great impact, good preprocessing of the collected data is an essential step in designing a good algorithm. Wavelet threshold denoising, Butterworth filtering and other means can achieve better smoothing effect, setting an ideal wavelet coefficient denoising threshold, or setting an appropriate Butterworth critical frequency can effectively solve the power frequency interference and For high-frequency noise problems, these parameters should be determined and improved according to the sampling frequency of the system and the trigger frequency of the effective signal. For the baseline drift, the present invention adopts the asymmetric least squares smoothing method to correct, and its asymmetry parameter and smoothness parameter depend on the situation. Figure 4 and Figure 5 show the system's preprocessing process for typical signals. Figure 4 shows the preprocessing effect of the breathing signal in the resting state, and Figure 5 shows the signal preprocessing effect when the fetal movement is triggered.

In step 2) of this embodiment, the batch of multi-channel pressure sensing data is cut into segments to form a single independent determination unit. The time span of a single determination unit needs to be determined by comprehensively considering the fetal movement trigger time and the system sampling frequency. Generally speaking , the time span of a single determination unit can be 2-15 seconds. If the determination unit within this time span has fetal movement characteristics, it is determined as "true", that is, fetal movement signal, otherwise it is determined as "false", that is, noise signal. Taking the 4-channel sensing data as an example, in the order of collection time, the 4-channel pressure sensing data is cut into segments to form multiple judgment units connected in sequence, and each judgment unit includes 4 channels collected in the same time segment. data.

In step 3) of this embodiment, the collected data is specifically classified into: percussion fetal movement, strong fetal movement, breathing noise, sneezing noise and body movement noise. Taking 4-channel sensing data as an example, the waveforms of several representative signals are listed as follows (as shown in Figure 6). The 4 curves in the figure represent the pressure data collected by the 4 sensing channels respectively. Figure 6a shows the data collected by the 4-channel sensor when there is no fetal movement. At this time, the 4-channel data is the abdominal wall pressure signal generated by the normal breathing of the pregnant woman, and each channel has obvious breathing signal characteristics; Figure 6b is a "tapping" form of The data waveform generated by fetal movement, the curve channel indicated by the arrow in the figure has obvious pressure pulse signal; Figure 6c is the data waveform generated by the "strong" fetal movement, the "strong" fetal movement can trigger multiple abdominal wall areas, the arrow in the figure The two curve channels referred to can monitor the obvious pressure jump process; Figure 6d shows the data waveform generated by the "sneezing" noise. The 4-channel data shows a significant pressure increase, and then quickly drops to around 0, and then returns to In the normal pressure data range, there are fewer regular peaks and troughs; Figure 6e shows the data waveform generated by "body motion" noise. The pressure average of the 4 channels will increase significantly at the same time, and slowly return to the normal value, and there is no regularity. Peaks and valleys. The algorithms provided in some embodiments of the present invention can perform feature fusion on the multi-channel sensor data of each determination unit, using shallow features (including but not limited to time-domain features, frequency-frequency features, wavelet series, etc.) and similarity Algorithms (including but not limited to Pearson correlation coefficient, Hausdorff distance, Frechet distance) and AI algorithms (including but not limited to KNN, logistic regression, SVM, etc.) Two labels of "true" and "false" are indicated), and further determine what form of fetal movement and what form of noise it belongs to.

In a preferred embodiment, the acquisition units include four uniformly spaced laterally arranged on the abdominal wall of the pregnant woman, so as to perform parallel acquisition of 4-channel data.

In step 3), the specific steps of analyzing and determining the determination unit by multi-channel feature analysis are:

3-1) Perform multi-channel feature extraction on a single decision unit, including:

3-1-1) Extract the Pearson correlation coefficient between the 4 channels. The calculation formula of the Pearson correlation coefficient is:

ρ _X,Y =cov(X,Y)/σ _X σ _Y =E((X-μ _X )(Y-μ _Y ))/σ _X σ _Y ;

Among them, cov(X, Y) represents the covariance between two variables X, Y, σ _X , σ _Y represents the standard deviation of the two variables X, Y; s0, s1, s2, s3 represent the four-channel passing through For the preprocessed vibration signal data, there are 6 Pearson correlation coefficients between the four channels, namely ρ _0,1 , ρ _0,2 , ρ _0,3 , ρ _1,2 , ρ _1,3 , ρ _{2 , 3.} Calculate the mean of the Pearson correlation coefficient of each channel for other channels, namely:

ρ ₀ =1/3(ρ _0,1 +ρ _0,2 +ρ _0,3 );

ρ ₁ =1/3(ρ _0,1 +ρ _1,2 +ρ _1,3 );

ρ ₂ =1/3(ρ _0,2 +ρ _1,2 +ρ _2,3 );

ρ ₃ =1/3(ρ _0,3 +ρ _1,3 +ρ _2,3 ).

Among them, the fetal movement signal is _a weak physiological parameter signal. Generally speaking, the _four pressure sensors will not be triggered _at the same time, and the Pearson correlation coefficient between the _four channels is obviously weaker. However, the noise signals such as breathing, sneezing, and body movement will stimulate the four pressure sensors at the same time, and the waveforms have similar development trends. Accordingly, they have high Pearson correlation coefficients, ρ ₀ , ρ ₁ , ρ ₂ , ρ ₃ is larger. The Pearson correlation coefficient ρ>=0.6 is strongly correlated, 0.4=<ρ<=0.6 is moderately correlated, and ρ<0.4 is weak or irrelevant.

3-1-2) Calculate the peak-to-peak quotient of each of the 4 channels: According to the sampling frequency, set the minimum horizontal unit distance between the peak and peak values, and at this distance, calculate the maximum peak value and the minimum peak value in each channel The ratio is recorded as peakDivision ₀ , peakDivision ₁ , peakDivision ₂ , and peakDivision ₃ .

3-1-3) Perform spectrum analysis and normalization on the four-channel signal, and calculate the average value amp _n of the corresponding amplitudes of the four channel signals at different frequencies, and count the number of channels whose amp _n is greater than the amplitude threshold thresholdAmp, denoted as count _{( amp>thresholdAmp)} ; Generally speaking, the normalized amplitude threshold thresholdAmp is about 0.2. The periodic regular breathing signal is expressed as a cosine signal, the main frequency count _{(amp>thresholdAmp)} is 2, and the two frequencies correspond to the constant term and the cosine frequency respectively.

3-1-4) Matrix feature analysis: In a single determination unit, the vibration signals of the 4 channel signals are formed into an M*4 matrix, where M represents the product of the sampling frequency f and the determination unit duration t, that is, M=f*t; : The sampling frequency is f=20Hz, and the duration of a single judgment unit is t=10s, then the pressure data of the four-channel signal forms a 200*4 matrix, as shown in Figure 7.

For this matrix, meanRowValue represents the mean value of the row vector, and index represents the index of the row vector, and the following characteristics are taken:

a) The difference between the maximum value and the minimum value of the row vector mean value r _max-min ,

r _max-min = meanRowValue _max -meanRowValue _min ,

Among them, meanRowValue _max represents the maximum value of the mean value of the row vector, and meanRowValue _min represents the minimum value of the mean value of the row vector;

b) The index difference between the index of the maximum value of the row vector mean and the index of the minimum value of the row vector mean index _max-min ,

Among them, index _max represents the index of the maximum value of the row vector mean value,

The index of the minimum value of the mean value of the row vector;

Among them, when r _max-min is greater than a certain threshold thresholdValue and the index difference index _max-min <1.5f, the sneeze signal can be effectively distinguished. Generally speaking, the threshold thresholdValue can be set to 3 times the breathing amplitude.

3-2) carry out the concrete classification of fetal movement signal and noise signal to the vibration signal of this judgment unit according to the multi-channel feature obtained in step 3-1), with reference to Fig. 8, including:

3-2-1) Set two Pearson correlation coefficient judgment thresholds A and B, where A=0.4, B=0.6; and compare the minimum value ρ _min in the mean value ρ _n of the Pearson correlation coefficient of all channels with A Compare with B, and then judge according to the following different steps according to the comparison result;

3-2-2) (ρ ₀ , ρ ₁ , ρ ₂ , ρ ₃ ) when _min > 0.6, count the peaks of each of the 4 channels

peak quotient peakDivisionN;

1. If there is a peak-to _- peak quotient of at least one channel greater than the set threshold PT, _PT = 2, then it is judged that the fetal movement in the form of percussion, the output classification result, otherwise continue to judge according to the following steps;

According to statistical experience, the Pearson correlation coefficient range is divided as follows: 0.8-1.0, very strong correlation; 0.6-0.8, strong correlation; 0.4-0.6, moderate correlation; 0.2-0.4, weak correlation; 0.0-0.2, very strong correlation Weak correlation or no correlation; <0, negative correlation; A and B two thresholds, need to comprehensively consider the number of channels N and the above statistical experience to determine and improve; in this embodiment, select P _T =2;

II. Count the number of count (amp>thresholdAmp ₎ , if count _{(amp>thresholdAmp)} = J, J = 2, then it is judged as breathing noise, and the classification result is output; otherwise, continue to judge according to the following steps; based on frequency domain analysis formula, the breathing signal is a cosine waveform, and the number of main frequencies is J=2;

III. Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is determined as sneezing noise, otherwise it is determined as body motion noise; output the classification result, where I=1.5; According to the experimental results, the threshold value thresholdValue needs to be determined and perfected according to the sensing characteristics of the sensing device (response curve, linearity, etc.), and the proportional coefficient I needs to be determined and perfected according to the sample performance. Generally speaking, 1≤I≤2.5; In this embodiment, I=1.5 is selected;

3-2-3) When ρ _min <A, if the number of Pearson correlation coefficients less than A is not greater than k in the Pearson correlation coefficients of all channels, it is determined as percussion fetal movement, otherwise it is determined as strong fetal movement , output the classification result; where k=3; according to the experimental results, the k value needs to be determined and perfected according to the number of channels N and the performance of the sample, in general, k>N/2; k=3 is selected in this embodiment;

3-2-4) When A < ρ _min < B, then count the peak-to-peak quotient peakDivisionN of each channel in the N channels. If there is at least one channel whose peak-to-peak quotient is greater than the set threshold 2, press again. The following steps a) are judged, otherwise the following steps b) are judged;

a) If among the Pearson correlation coefficients of all channels, the number of Pearson correlation coefficients less than A is not greater than k, it is determined as a percussion form of fetal movement, otherwise it is determined as a strong form of fetal movement, and the classification result is output; wherein k=3;

b) Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is judged as sneezing noise, otherwise it is judged as body motion noise; output the classification result, where I=1.5.

In this embodiment, the signals are classified into five categories: noise (breathing), noise (body movement), noise (sneezing), fetal movement (knocking), and fetal movement (strong) through the above algorithm, and labels 0 (breathing), -1 (body motion), -2 (sneezing), 1 (tapping), 2 (strong), as shown in Figure 7. In this embodiment, the performance of the classification algorithm is also evaluated, and the method used is: calculating the precision and recall indexes of each class, and multiplying by the weight proportion of the class in the total number of samples, The precision and recall indicators of the entire classification model are obtained, and the F1 value of the entire classification model is derived to determine the performance of the algorithm: the higher the F1 value, the more ideal the classification model is.

accuracy

recall

Precision=precision ₀ *w ₀ +precision _-2 *w _-2 +precision _-1 *w _-1 +precision ₁ *w ₁ +precision ₂ *w ₂ (3)

Recall=recall ₀ *w ₀ +recall _-2 *w _-2 +recall _-1 *w _-1 +recall ₁ *w ₁ +recall ₂ *w ₂ (4)

In the above formulas, (1) and (2) represent the algorithms for precision and recall, respectively, TP means that the correct prediction is a positive example, FP means that the wrong prediction is a positive example, and FN means that the wrong prediction is a negative example. The meaning of precision is: the proportion of correctly predicted positive (TP) accounts for all predicted positives (TP+FP), and the meaning of recall is: correctly predicted positive (TP) accounts for all positive samples (TP) +FN) ratio;

(3) and (4) represent the calculation formulas of Precision and Recall of the design algorithm of the present invention, w ₀ represents the weight of label 0 (breath) in all samples, w _-2 , w _-1 , w ₁ , w ₂ And so on; precision ₀ indicates the precision rate of the class identified as label 0 (breath), recall ₀ indicates the recall rate of the class identified as label 0 (breath), and so on for other categories;

(5) Represents the calculation formula of the F1 value of the algorithm of the present invention.

evaluation result:

Randomly select the fetal movement monitoring data of a pregnant woman, taking the pregnancy days of a pregnant woman as 30 weeks and 2 days, 32 weeks and 7 days, and 36 weeks and 4 days, a total of 1498 groups of data as an example. The horizontal axis represents the true class, and the vertical axis represents the predicted class. The following takes label 0 (breathing) as an example to illustrate: there are actually 949 cases of label 0 (breathing) in the test set, the first row and the first column indicate: the actual label 0 (breathing), the predicted label 0 (breathing) has a total of 880 Example; the first row and the second column indicate: the actual label is 0 (breathing), and there are 0 cases predicted to be label-2 (sneezing); the first row and the third column indicate: the actual label 0 (breathing), the predicted label is There are 0 cases of -1 (body movement); the first row and the fourth column indicate: the actual label is 0 (breathing), and there are 50 cases predicted to be label 1 (tapping); the first row and the fifth column indicate: the actual There are 19 cases with label 0 (breathing) predicted to be label 2 (strong).

According to the above evaluation method, after weight processing, it is calculated that the accuracy of the classification model of this algorithm is: 0.920, the recall rate is: 0.912, and the F1 value is: 0.916. Fetal movement determination belongs to weak physiological signal monitoring, and there is a lot of noise interference. This classification method achieves such performance while taking into account the determination of fetal movement form and the identification of noise types, which fully shows that this classification method has excellent performance.

Table 1

actual/forecast		Breath 0	sneeze-2	body movement-1	tap 1	intense 2
Breath 0	880	0	0	50	19
sneeze-2	0	16	1	0	0
body movement-1	3	0	98	9	7
tap 1	twenty two	0	0	294	10
intense 2	1	0	3	7	78

Although the embodiment of the present invention has been disclosed as above, it is not limited to the application listed in the description and the embodiment, and it can be applied to various fields suitable for the present invention. For those skilled in the art, it can be easily Therefore, the invention is not limited to the specific details without departing from the general concept defined by the appended claims and the scope of equivalents.

Claims (10)

1. An intelligent wearable fetal movement monitoring system, comprising: a data acquisition module, an intelligent terminal, a data storage and analysis and determination module and a data management module deployed on a cloud server;

   The data acquisition module includes N acquisition units arranged on the abdominal wall of the pregnant woman at intervals for collecting vibration signals of the abdominal wall of the pregnant woman, so as to perform parallel acquisition of N channel data;

   The intelligent terminal is used for transmitting the multi-channel data collected by the data collection module to the data storage and analysis and determination module;

   The data storage and analysis and determination module includes a data storage unit and an analysis and determination unit, the data storage unit receives and stores the multi-channel data sent by the intelligent terminal, and the analysis and determination unit analyzes and determines the multi-channel data to The vibration signal distinguishes the fetal movement signal and the noise signal, and further distinguishes and determines the specific form of the fetal movement signal and the noise signal to form a classification result; wherein, the specific form of the fetal movement signal includes at least the percussion form of fetal movement and the strong form of fetal movement, and the noise signal. The specific forms of noise include at least breathing noise, sneezing noise and body motion noise;

   The data management module includes a statistical unit and an application terminal, and the data management module receives the classification result sent by the data storage and analysis and determination module, and accordingly forms a multi-dimensional fetal movement index per unit monitoring time;

   The application terminal displays the multi-dimensional fetal movement index obtained by the statistical unit.
2. The smart wearable fetal movement monitoring system according to claim 1, wherein the acquisition unit is a wearable pressure sensor arranged on the abdominal wall of the pregnant woman.
3. The smart wearable fetal movement monitoring system according to claim 1, wherein the application terminal includes a management background and a personal user terminal.
4. The intelligent wearable fetal movement monitoring system according to claim 3, wherein the multi-dimensional fetal movement indicators at least include the number of fetal movements in a unit monitoring time, the form of fetal movement, the longest duration of fetal movement, and the strong fetal movement in the longest duration. Proportion, longest fetal static time and fetal activity;

   Wherein, the longest duration of the fetal movement is: the total time occupied by the interval with the longest duration among the determination unit intervals that are continuously determined as fetal movement signals within the unit monitoring time;

   In the described longest duration, the proportion of strong fetal movement is: in the unit monitoring time, in the longest duration interval of fetal movement, the percentage that is judged as the interval of strong form fetal movement accounts for;

   The longest static time of the fetus is: the total time occupied by the interval with the longest duration among the determination unit intervals that are continuously determined to be noise signals within the unit monitoring time;

   The fetal activity is: the percentage of the determination units determined as fetal movement signals in all determination units in the unit monitoring time in the unit monitoring time.
5. The intelligent wearable fetal movement monitoring system according to claim 4, wherein the method for performing fetal movement monitoring by the intelligent wearable fetal movement monitoring system comprises the following steps:

   S1, carry out the parallel collection of the vibration signal data of N channels to the abdominal wall of pregnant women by the data collection module;

   S2. The intelligent terminal wirelessly transmits the multi-pass data collected by the data collection module to the data storage and analysis and determination module;

   S3. The data storage unit in the data storage and analysis and determination module receives and stores the multi-channel data sent by the intelligent terminal, and then the analysis and determination unit in the data storage and analysis and determination module analyzes and determines the multi-channel data, The vibration signal is distinguished from the fetal movement signal and the noise signal, and the specific form of the fetal movement signal and the noise signal is further distinguished and determined to form a classification result;

   S4, the data management module receives the classification result sent by the data storage and analysis and determination module, and accordingly forms a multi-dimensional fetal movement index per unit monitoring time;

   S5. The management background and the personal user terminal in the application terminal display the obtained multi-dimensional fetal movement index for the user to obtain.
6. The smart wearable fetal movement monitoring system according to claim 5, wherein the method for analyzing and determining the multi-channel data by the analyzing and determining unit in the step S3 comprises the following steps:

   1) Data preprocessing;

   2) For the preprocessed data, firstly, the data of the N channels received in the unit monitoring time is cut and segmented according to the timeline to form multiple independent judgment units, and then the multiple judgment units are divided one by one according to the timeline. Analyze and determine through the following steps;

   3) Through multi-channel feature analysis, the vibration signal is firstly judged to distinguish the fetal movement signal and the noise signal, and then the specific form of the fetal movement signal and the noise signal is discriminated and judged to form a classification result, and then the analysis and judgment of the next judgment unit is carried out;

   4) Repeat the above step 3) until the analysis and determination of all the determination units within the unit monitoring time are completed.
7. The smart wearable fetal movement monitoring system according to claim 6, wherein the step 1) specifically includes: firstly, the vibration signal is smoothed by a method including at least wavelet threshold denoising and Butterworth filtering, The baseline drift correction was then performed using the asymmetric least squares baseline correction method.
8. The smart wearable fetal movement monitoring system according to claim 6, wherein the step 3) specifically comprises:

   3-1) Perform multi-channel feature extraction on a single decision unit, including:

   3-1-1) Extract the Pearson correlation coefficient between the N channels, then calculate the mean value of the Pearson correlation coefficient for each channel for other channels, and denote it as the Pearson correlation coefficient mean value ρ _n ;

   3-1-2) Calculate the peak-to-peak quotient of each channel in the N channels: According to the sampling frequency, set the minimum horizontal unit distance between peak-to-peak values, and at this distance, calculate the maximum peak value and minimum peak value of each channel The ratio, denoted as peakDivision _N ;

   3-1-3) Perform spectrum analysis and normalization on the N channel signals, and calculate the average value amp _n of the corresponding amplitudes of the N channel signals at different frequencies, and count the number of channels whose amp _n is greater than the amplitude threshold thresholdAmp, denoted as count _{(amp>thresholdAmp)} ;

   3-1-4) Matrix feature analysis: In a single judgment unit, the vibration signal data of N channel signals are formed into an M*N matrix, where M represents the product of the sampling frequency f and the judgment unit duration t, that is, M=f*t ; For this matrix, meanRowValue represents the mean value of the row vector, and index represents the index of the row vector, taking the following characteristics:

   a) The difference r _max-min between the row vector mean maximum value and the minimum value, r _max-min = meanRowValue _max - meanRowValue _min , wherein meanRowValue _max represents the row vector mean maximum value, and meanRowValue _min represents the row vector mean mean minimum value;

   b) The index difference between the index of the maximum value of the row vector mean and the index of the minimum value of the row vector mean index _max-min ,

   

   Among them, index _max represents the index of the maximum value of the row vector mean value,

   

   The index of the minimum value of the mean value of the row vector;

   2-2) carry out the concrete classification of fetal movement signal and noise signal to the vibration signal of this determination unit according to the multi-channel feature obtained in step 2-1), including:

   3-2-1) Set two Pearson correlation coefficient judgment thresholds A and B, where 0<A<B, and compare the minimum value ρ _min in the mean value ρ _n of the Pearson correlation coefficient of all channels with A and B. Compare, and then judge according to the comparison results according to the following different steps;

   3-2-2) When ρ _min >B, count the peak-to-peak quotient peakDivisionN of each channel in the N channels;

   1. If the peak-to-peak quotient of at least one channel is greater than the set threshold P _T , it is determined to be a fetal movement in the form of percussion, and the classification result is output, otherwise, the determination is continued according to the following steps, wherein, 2.0≤P _T ≤2.5 ;

   II. Count the number of count _{(amp>thresholdAmp)} , if the number of main frequencies count _{(amp>thresholdAmp)} =J, then judge as breathing noise, output the classification result, J is a positive integer; otherwise, continue to judge according to the following steps;

   III. Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is determined as sneezing noise, otherwise it is determined as body motion noise; output the classification result, where I is the proportional coefficient , 1≤I≤2.5;

   3-2-3) When ρ _min <A, if the number of Pearson correlation coefficients less than A is not greater than k in the Pearson correlation coefficients of all channels, it is determined as percussion fetal movement, otherwise it is determined as strong fetal movement , output the classification result; where k>N/2;

   3-2-4) When A < ρ _min < B, then count the peak-to-peak quotient peakDivisionN of each channel in the N channels, if there is at least one channel whose peak-to-peak quotient is greater than the set threshold P _T , then count again. Determine according to the following step a), otherwise according to the following step b) to determine;

   a) If among the Pearson correlation coefficients of all channels, the number of Pearson correlation coefficients less than A is not greater than k, it is determined as a percussion form of fetal movement, otherwise it is determined as a strong form of fetal movement, and the classification result is output; where k>N/ 2;

   b) Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is judged as sneezing noise, otherwise it is judged as body motion noise; output the classification result, where I is the proportional coefficient , 1≤I≤2.5.
9. The intelligent wearable fetal movement monitoring system according to claim 8, characterized in that, the acquisition unit includes four units arranged on the abdominal wall of pregnant women at uniform intervals along the lateral direction, so as to perform parallel acquisition of 4-channel data.
10. The smart wearable fetal movement monitoring system according to claim 9, wherein the step 3) specifically comprises:

    3-1) Perform multi-channel feature extraction on a single decision unit, including:

    3-1-1) Extract the Pearson correlation coefficient between the 4 channels. The calculation formula of the Pearson correlation coefficient is:

    ρ _X,Y =cov(X,Y)/σ _X σ _Y =E((X-μ _X )(Y-μ _Y ))/σ _X σ _Y ;

    Among them, cov(X, Y) represents the covariance between two variables X, Y, σ _X , σ _Y represents the standard deviation of the two variables X, Y; s0, s1, s2, s3 represent the four-channel passing through For the preprocessed vibration signal data, there are 6 Pearson correlation coefficients between the four channels, namely ρ _0,1 , ρ _0,2 , ρ _0,3 , ρ _1,2 , ρ _1,3 , ρ _{2 , 3.} Calculate the mean of the Pearson correlation coefficient of each channel for other channels, namely:

    ρ ₀ =1/3(ρ _0,1 +ρ _0,2 +ρ _0,3 );

    ρ ₁ =1/3(ρ _0,1 +ρ _1,2 +ρ _1,3 );

    ρ ₂ =1/3(ρ _0,2 +ρ _1,2 +ρ _2,3 );

    ρ ₃ =1/3(ρ _0,3 +ρ _1,3 +ρ _2,3 );

    3-1-2) Calculate the peak-to-peak quotient of each of the 4 channels: According to the sampling frequency, set the minimum horizontal unit distance between the peak and peak values, and at this distance, calculate the maximum peak value and the minimum peak value in each channel The ratio, denoted as peakDivision ₀ , peakDivision ₁ , peakDivision ₂ , peakDivision ₃ ;

    3-1-3) Perform spectrum analysis and normalization on the four-channel signal, and calculate the average value amp _n of the corresponding amplitudes of the four channel signals at different frequencies, and count the number of channels whose amp _n is greater than the amplitude threshold thresholdAmp, denoted as count _{( amp>thresholdAmp)} ;

    3-1-4) Matrix feature analysis: In a single determination unit, the vibration signals of the 4 channel signals are formed into an M*4 matrix, where M represents the product of the sampling frequency f and the determination unit duration t, that is, M=f*t; In this matrix, meanRowValue represents the mean value of the row vector, and index represents the index of the row vector, and takes the following characteristics:

    a) The difference between the maximum value and the minimum value of the row vector mean value r _max-min ,

    r _max-min = meanRowValue _max -meanRowValue _min ,

    Among them, meanRowValue _max represents the maximum value of the mean value of the row vector, and meanRowValue _min represents the minimum value of the mean value of the row vector;

    b) The index difference between the index of the maximum value of the row vector mean and the index of the minimum value of the row vector mean index _max-min ,

    

    Among them, index _max represents the index of the maximum value of the mean value of the row vector,

    

    The index of the minimum value of the mean value of the row vector;

    3-2) carry out the concrete classification of fetal movement signal and noise signal to the vibration signal of this judgment unit according to the multi-channel feature obtained in step 3-1), including:

    3-2-1) Set two Pearson correlation coefficient judgment thresholds A and B, where A=0.4, B=0.6; and compare the minimum value ρ _min in the mean value ρ _n of the Pearson correlation coefficient of all channels with A Compare with B, and then judge according to the following different steps according to the comparison result;

    3-2-2) When (ρ ₀ , ρ ₁ , ρ ₂ , ρ ₃ ) _min > 0.6, count the peak-to-peak quotient peakDivisionN of each of the 4 channels;

    1. If there is a peak-to _- peak quotient of at least one channel greater than the set threshold PT, _PT = 2, then it is judged that the fetal movement in the form of percussion, the output classification result, otherwise continue to judge according to the following steps;

    II. Count the number of count (amp>thresholdAmp ₎ , if count _{(amp>thresholdAmp)} =J, J=2, then it is judged as breathing noise, and the classification result is output; otherwise, continue to judge according to the following steps;

    III. Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is determined as sneezing noise, otherwise it is determined as body motion noise; output the classification result, where I=1.5;

    3-2-3) When ρ _min <A, if the number of Pearson correlation coefficients less than A is not greater than k in the Pearson correlation coefficients of all channels, it is determined as percussion fetal movement, otherwise it is determined as strong fetal movement , output the classification result; where k=3;

    3-2-4) When A < ρ _min < B, then count the peak-to-peak quotient peakDivisionN of each channel in the N channels. If there is at least one channel whose peak-to-peak quotient is greater than the set threshold 2, press again. The following steps a) are judged, otherwise the following steps b) are judged;

    a) If among the Pearson correlation coefficients of all channels, the number of Pearson correlation coefficients less than A is not greater than k, it is determined as a percussion form of fetal movement, otherwise it is determined as a strong form of fetal movement, and the classification result is output; wherein k=3;

    b) Set the matrix feature analysis threshold thresholdValue, if r _max-min >thresholdValue and index _max-min <I*f, it is judged as sneezing noise, otherwise it is judged as body motion noise; output the classification result, where I=1.5.

Images