A real-time fetal heart sound monitoring system based on multi-microphone array

Abstract

The present invention relates to the technical field of fetal heart sound monitoring, and specifically to a real-time fetal heart sound monitoring system based on a multi-microphone array. The present invention establishes relative coordinates by monitoring target positions with multiple microphones, and determines the data type of physiological signal interference obtained by each microphone based on the coordinates; through a fetal heart sound stethoscope analysis model, it is possible to perform similarity comparison on the data corresponding to the interference data type of each monitored fetal heart sound data according to the interference data type obtained by each microphone; at the same time, based on the existing model, the monitored fetal heart sound data and abnormal fetal heart sound data are compared and analyzed; through this method, based on the large amount of data from the microphone array, the amount of data for noise reduction work can be minimized to the greatest extent, and the physiological signal interference can be eliminated to the greatest extent, thereby ensuring the accuracy of the monitored fetal heart sound data; at the same time, it can also provide a data basis for subsequent fetal heart sound analysis based on the existing model, thereby improving the quality of abnormal fetal heart sound data analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of fetal heart sound monitoring, and in particular to a real-time fetal heart sound monitoring system based on a multi-microphone array. Background Art

[0002] Fetal heart rate monitoring technology is an important monitoring method that monitors the health of the fetus during pregnancy by acquiring the sounds of the fetal heart activity inside the pregnant woman's body. Traditionally, fetal heart rate monitoring methods have used a stethoscope. With the development of technology, the most commonly used monitoring method today is Doppler fetal heart rate monitoring, which uses ultrasound to monitor fetal heart rates. This method collects ultrasound data, visualizes it, and stores it. Compared to the monitoring results of the stethoscope method, the monitoring accuracy of Doppler fetal heart rate monitoring has been significantly improved.

[0003] However, Doppler fetal heart rate monitors also have many shortcomings. In actual use, there is still a lot of environmental signal noise interference, which will affect the monitoring results of the Doppler fetal heart rate monitor. At the same time, the Doppler fetal heart rate monitor also faces the problem of inaccurate fetal heart rate data acquisition due to the location problem of obtaining fetal heart rate data.

[0004] In order to solve the above problems, fetal heart sound monitoring with multi-microphone arrays can, to a certain extent, solve the shortcomings of Doppler fetal heart monitors. On the one hand, a large number of microphone arrays can filter out data with higher signal-to-noise ratio from multiple sets of data for analysis. On the other hand, the acquisition of a large amount of data in different directions can also reduce the impact of monitoring results caused by different monitoring positions. However, the current monitoring method of multi-microphone arrays still has some challenges. Since a large amount of data is acquired, the screening and troubleshooting of data is more cumbersome. Although current signal processing technology can process some noise, there are still some problems with some regular noise, such as physiological signal noise such as bowel sounds and heartbeats of pregnant women.

[0005] In order to improve the quality of fetal heart sound data monitoring using a multi-microphone array, a real-time fetal heart sound monitoring system based on a multi-microphone array was proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a real-time fetal heart sound monitoring system based on a multi-microphone array, which establishes relative coordinates by monitoring the target position through multiple microphones, and determines the data type of physiological signal interference obtained by each microphone based on the coordinates; through the fetal heart sound stethoscope analysis model, it can perform similarity comparison on the data corresponding to the interference data type of each monitored fetal heart sound data according to the interference data type obtained by each microphone, and eliminate physiological signal interference to the greatest extent; at the same time, based on the existing model, the monitored fetal heart sound data and abnormal fetal heart sound data are compared and analyzed; through this method, based on the large amount of data from the microphone array, the amount of data for noise reduction work can be minimized, and the physiological signal interference can be eliminated to the greatest extent, ensuring the accuracy of the monitored fetal heart sound data, and at the same time, it can provide a data basis for subsequent fetal heart sound analysis based on the existing model, and improve the quality of abnormal fetal heart sound data analysis.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A real-time fetal heart sound monitoring system based on a multi-microphone array includes a fetal heart sound stethoscope array position module, a fetal heart sound stethoscope data module, a fetal heart sound stethoscope interference analysis module, a fetal heart sound stethoscope model analysis module, and a fetal heart sound stethoscope model abnormality module, specifically:

[0009] a fetal heart sound stethoscope array position module, which obtains the stethoscope position of the stethoscope target, establishes a relative stethoscope coordinate system based on the stethoscope position, and determines the positions of multiple microphone stethoscopes based on the central stethoscope position of the multi-microphone array;

[0010] Furthermore, the stethoscope target includes a plurality of stethoscope levels, and a plurality of relative stethoscope coordinate centers are obtained according to the stethoscope levels;

[0011] Furthermore, a stethoscope microphone coordinate system is established with the central stethoscope position, and the microphone stethoscope coordinates of each microphone are obtained; the microphone coordinate system is a proportional mapping of the relative stethoscope coordinate system;

[0012] According to the relative stethoscope coordinate system, the relative stethoscope center coordinates of the central stethoscope position are obtained; according to the relative stethoscope center coordinates and the microphone stethoscope coordinates, the relative stethoscope center coordinates of each microphone are determined, and the relative stethoscope center coordinates of each microphone are the microphone stethoscope position;

[0013] a fetal heart sound stethoscope data module, which acquires multiple sets of first stethoscope data at different stethoscope positions based on the multi-microphone array;

[0014] a fetal heart sound stethoscope interference analysis module, which obtains an interference stethoscope target for each set of the first stethoscope data based on the plurality of sets of the first stethoscope data and the relative stethoscope position coordinates of the microphone stethoscope position; and extracts corresponding second stethoscope data based on the interference stethoscope target;

[0015] Furthermore, the relative stethoscope coordinates of each interference target are obtained based on the relative stethoscope coordinate center of the stethoscope level; the interference stethoscope target of each group of the first stethoscope data is determined based on the interference stethoscope distance of the relative stethoscope center coordinates of each microphone and the relative stethoscope coordinates of each interference target; a plurality of interference stethoscope weights are divided according to the ratio of the interference stethoscope distance and the interference stethoscope threshold, and whether it is an interference stethoscope target is determined according to the interference stethoscope weight;

[0016] a fetal heart sound stethoscope model analysis module, which analyzes first stethoscope data features of the first stethoscope data using a fetal heart sound stethoscope analysis model, and analyzes second stethoscope data features based on the second stethoscope data; performs a stethoscope data feature comparison between the first stethoscope data features and the second stethoscope data features, and performs stethoscope interference processing on the first stethoscope data having the same features as the second stethoscope data; the first stethoscope data is fetal heart sound monitoring data, and the second stethoscope data is fetal heart sound monitoring interference data;

[0017] a fetal heart sound stethoscope model abnormality module, which obtains third stethoscope data features of third stethoscope data through the fetal heart sound stethoscope analysis model, compares the third stethoscope data features with the first stethoscope data features after stethoscope interference processing, and analyzes abnormal fetal heart sound data, wherein the third stethoscope data is abnormal fetal heart sound data;

[0018] Furthermore, the first stethoscope data feature, the second stethoscope data feature, and the third stethoscope data feature are acquired by a stethoscope feature data extraction module of the fetal heart sound stethoscope analysis model; the stethoscope feature data extraction module includes a stethoscope data noise reduction preprocessing unit, a stethoscope data segmentation unit, and a stethoscope data feature extraction storage unit;

[0019] The stethoscope data denoising preprocessing unit performs denoising on the input stethoscope data; preprocesses the denoised input stethoscope data to generate feature-to-be-extracted stethoscope data; the input stethoscope data includes the first stethoscope data, the second stethoscope data and the third stethoscope data;

[0020] The stethoscope data segmentation unit performs data segmentation on the stethoscope data to be extracted with features to generate segmented stethoscope data with features to be extracted;

[0021] Further, a plurality of corresponding second stethoscope data are obtained according to the plurality of interfering stethoscope targets;

[0022] Calculating the average frequency of all the second stethoscope data corresponding to each of the interfering stethoscope targets; calculating the average frequency of all the third stethoscope data;

[0023] Generate a corresponding stethoscope interference label for each of the interfering stethoscope targets and associate it with the corresponding average frequency; segment the second stethoscope data according to the average frequency;

[0024] Obtaining the average frequency of the corresponding stethoscope interference label according to the interference stethoscope weight, and segmenting the first stethoscope data in combination with the average frequency; the segmented first stethoscope data includes multiple groups;

[0025] Calculating the maximum frequency of all the third stethoscope data; segmenting the third stethoscope data according to the maximum frequency; and segmenting the first stethoscope data in combination with the maximum frequency;

[0026] The stethoscope data feature extraction unit extracts the stethoscope feature data based on the segmentation feature to be extracted stethoscope data, and performs data association storage; the stethoscope data feature extraction unit includes 2 layers of Transformer layers, 4 layers of LSTM layers, 2 layers of residual convolution layers, 1 layer of feature fusion layer and a classification layer;

[0027] Furthermore, the first stethoscope data feature is compared with the second stethoscope data feature, and the first stethoscope data feature is compared with the third stethoscope data feature, and similarity is calculated by the stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model, and the stethoscope data feature comparison module is:

[0028] σ S =α1·f cos (l in ,l S )+α2·f D (l in ,l S )+α3·f F (l in ,l S )+α4·f B (l in ,l S );

[0029] Among them, σ S is the similarity result, f cos is the cosine similarity, f D is the dynamic time planning similarity, f Fis the Fourier similarity, f B is the empirical statistical similarity, l in is the first stethoscope data feature, l S is the data feature being compared, l S Including the second stethoscope data feature and the third stethoscope data feature, α1, α2, α3 and α4 are similarity calculation weights.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The present invention generates corresponding coordinates for calculation through the fetal position, pregnant woman position and multi-microphone array position at different periods, and matches the position information of the multi-microphone array with the position information of the monitored stethoscope target. Through this method, in the process of multi-microphone array monitoring of fetal heart sounds, it is possible to understand the interference caused by different physiological information that each microphone is susceptible to. Based on the calculated distance data, it is possible to screen out the factors that interfere with the fetal heart sounds of different microphones, so as to reduce the analysis work of data noise reduction, and provide an effective data basis for subsequent fetal heart sound data monitoring and data noise reduction.

[0032] 2. The present invention uses the filtered interference data, the stethoscope feature data extraction module and the stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model to extract the currently monitored fetal heart sound data and the stored interference noise data features, and performs automated feature comparison to perform data processing. This method can quickly calculate the interference noise data in the fetal heart sound data in a large amount of data and eliminate the data, so as to minimize the interference of the interference noise data and improve the monitoring quality of the fetal heart sound data.

[0033] 3. Based on the denoised data, the present invention uses the stethoscope feature data extraction module and the stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model to compare the abnormal fetal heart sound data with the monitored fetal heart sound data. By utilizing the characteristics of the original model, the fetal heart sound data can be monitored in addition to filtering out the noise data. This method can maximize the robustness of the fetal heart sound stethoscope analysis model, while ensuring the monitoring quality of the data, and can also achieve the purpose of monitoring abnormal fetal heart sound data, providing a reliable data basis for the monitoring process, and improving the data quality from the aspect of monitoring abnormal fetal heart sound data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a system flow chart of the present invention;

[0035] Figure 2 Schematic diagram of the microphone coordinate system of the present invention;

[0036] Figure 3A schematic diagram of the stethoscope microphone coordinate system and determination of microphone coordinates relative to the stethoscope coordinate system of the present invention;

[0037] Figure 4 A virtual device diagram of a stethoscope data denoising processing unit, a stethoscope data segmentation unit, a stethoscope data feature extraction and storage unit, and a stethoscope data feature comparison module of the stethoscope feature data extraction module of the fetal heart sound stethoscope analysis model of the present invention;

[0038] Figure 5 A network structure diagram of a stethoscope data feature extraction and storage unit of a fetal heart sound stethoscope analysis model of the present invention;

[0039] Figure 6 This is a graph of the loss function for identifying abnormal fetal heart sounds using the Transformer layer, four LSTM layers, and two residual convolution layers of the present invention.

[0040] Figure 7 This is a loss function diagram for the fetal heart sound stethoscope analysis model of the present invention to identify abnormal fetal heart sounds. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Fetal heart sound is the main method for monitoring fetal health before birth. Since fetal heart sound is easily affected by the position of the monitoring instrument and environmental factors, with the advancement of technology, multi-microphone arrays can reduce the interference of instrument position and environmental factors to a certain extent. However, this solution still has certain defects. Due to the large amount of data, the data processing work is relatively cumbersome; at the same time, this solution still has some problems in processing the physiological noise of some pregnant women. In order to improve the quality of monitoring data, the present invention provides a real-time fetal heart sound monitoring system based on a multi-microphone array, referring to Figure 1 As shown, the technical solution is as follows:

[0043] a fetal heart sound stethoscope array position module, which obtains the stethoscope position of the stethoscope target, establishes a relative stethoscope coordinate system based on the stethoscope position, and determines the positions of multiple microphone stethoscopes based on the central stethoscope position of the multi-microphone array;

[0044] a fetal heart sound stethoscope data module, which acquires multiple sets of first stethoscope data at different stethoscope positions based on the multi-microphone array;

[0045] a fetal heart sound stethoscope interference analysis module, which obtains an interference stethoscope target for each set of the first stethoscope data based on the plurality of sets of the first stethoscope data and the relative stethoscope position coordinates of the microphone stethoscope position; and extracts corresponding second stethoscope data based on the interference stethoscope target;

[0046] a fetal heart sound stethoscope model analysis module, which analyzes first stethoscope data features of the first stethoscope data using a fetal heart sound stethoscope analysis model, and analyzes second stethoscope data features based on the second stethoscope data; performs a stethoscope data feature comparison between the first stethoscope data features and the second stethoscope data features, and performs stethoscope interference processing on the first stethoscope data having the same features as the second stethoscope data;

[0047] The fetal heart sound stethoscope model abnormality module obtains the third stethoscope data feature of the third stethoscope data through the fetal heart sound stethoscope analysis model, compares the third stethoscope data feature with the first stethoscope data feature after stethoscope interference processing, and analyzes the abnormal fetal heart sound data.

[0048] To maximize the filtering of interference data, a coordinate system is established using the position of the pregnant woman and the multi-microphone array. Based on this coordinate system, the distance of each microphone relative to the pregnant woman can be calculated to determine the interference experienced by each microphone. This method can filter the interference data of each microphone before noise processing, reducing the workload of interference data processing and providing a reliable data foundation for subsequent improvement of fetal heart sound data quality.

[0049] Based on the interference data obtained by each microphone, the data characteristics currently monitored by each microphone are analyzed through the fetal heart sound stethoscope analysis model. The interference data characteristics corresponding to the data characteristics monitored by each microphone are obtained. The monitoring data characteristics are compared with the interference data characteristics to obtain their data similarity to distinguish whether the current monitoring data has interference data. This method can obtain the corresponding data analysis characteristics based on the screened interference targets for similarity matching, so as to distinguish whether the current monitoring data has interference and perform subsequent processing of the interference data, thereby improving the data quality of the fetal heart sound monitoring data.

[0050] At the same time, based on the existing model, the processed fetal heart sound monitoring data is compared with the abnormal fetal heart sound monitoring data. This method can maximize the utilization of the existing model, not only to analyze and improve data quality in the screening of data interference, but also to improve data quality in the health monitoring of fetal heart sound.

[0051] Example 1

[0052] For the purpose of specific description, the present invention will describe filtering interference content in conjunction with the following content:

[0053] A real-time fetal heart sound monitoring system based on a multi-microphone array includes a fetal heart sound stethoscope array position module, a fetal heart sound stethoscope data module, a fetal heart sound stethoscope interference analysis module, a fetal heart sound stethoscope model analysis module, and a fetal heart sound stethoscope model abnormality module, specifically:

[0054] a fetal heart sound stethoscope array position module, which obtains the stethoscope position of the stethoscope target, establishes a relative stethoscope coordinate system based on the stethoscope position, and determines the positions of multiple microphone stethoscopes based on the central stethoscope position of the multi-microphone array;

[0055] Furthermore, the stethoscope target includes multiple stethoscope levels, and multiple relative stethoscope coordinate centers are obtained based on the stethoscope levels; the coordinate center is the position center of the fetus, and the coordinates can be defined based on the world coordinate system of the pregnant woman when monitoring fetal heart sounds, or the monitoring room or monitoring instrument as a reference coordinate system to define the position center of the fetus; the position of the coordinate center is not unique and can be defined based on actual monitoring conditions;

[0056] The position of the fetus at different stages of pregnancy is estimated by combining a large amount of data, expert experience and modeling technology;

[0057] Because the fetus's position in the pregnant woman's body varies at different stages of pregnancy, its distance from the multi-microphone array also varies. Therefore, expert experience is combined to simulate position information at different stages of pregnancy and make estimates to determine the distance between the microphone and the fetus. This facilitates subsequent statistics of the multi-microphone array's distance to the fetus and the interference information received by different microphones, providing a data basis for subsequent calculations and improving the quality of fetal heart rate monitoring.

[0058] Furthermore, a stethoscope microphone coordinate system is established with the central stethoscope position, referring to Figure 2 As shown, and obtain the microphone stethoscope coordinates of each microphone; the microphone coordinate system is a proportional mapping of the relative stethoscope coordinate system; refer to Figure 3 As shown, Figure 3 The xyz coordinate system is relative to the stethoscope coordinate system. Figure 3 The x'y'z' in the figure is the stethoscope microphone coordinate system;

[0059] According to the relative stethoscope coordinate system, the relative stethoscope center coordinates of the central stethoscope position are obtained, and the relative stethoscope center coordinates of the central stethoscope position are obtained. Figure 3 According to the relative stethoscope center coordinates and the microphone stethoscope coordinates, the microphone stethoscope coordinates are A2 coordinates in the xyz coordinate system, determine the relative stethoscope center coordinates of each microphone, reference Figure 3As shown, the relative coordinates of each microphone to the center of the stethoscope are the coordinates of A3 in the xyz coordinate system, and the relative coordinates of each microphone to the center of the stethoscope are the positions of the microphones on the stethoscope;

[0060] Since the amount of data from a multi-microphone array is large, if the coordinates of each microphone relative to the set stethoscope center are calculated, the amount of data may be too large and the calculation accuracy may be affected. Since the position of the multi-microphone array is relatively fixed, the center of the microphone array is used as one of the reference positions to accurately calculate the position information of the microphone. The coordinate position of the microphone array center is determined by the coordinates of the microphone array center and the relative stethoscope center. The positions of the other microphones are determined by combining the positions of the microphone center and other microphones. This method can reduce a large number of variables and improve the calculation accuracy. It can also improve the calculation accuracy of each microphone array, facilitate the determination of possible interference targets for each microphone, and improve the quality of fetal heart sound monitoring.

[0061] a fetal heart sound stethoscope data module, which acquires multiple sets of first stethoscope data at different stethoscope positions based on the multi-microphone array;

[0062] a fetal heart sound stethoscope interference analysis module, which obtains an interference stethoscope target for each set of the first stethoscope data based on the plurality of sets of the first stethoscope data and the relative stethoscope position coordinates of the microphone stethoscope position; and extracts corresponding second stethoscope data based on the interference stethoscope target;

[0063] Furthermore, the relative stethoscope coordinates of each interference target are obtained based on the relative stethoscope coordinate center of the stethoscope level; the interference stethoscope target of each group of the first stethoscope data is determined based on the relative stethoscope center coordinates of each microphone and the interference stethoscope distance of the relative stethoscope coordinates of each interference target; through the determined multi-microphone array position, the actual environment can be digitized to obtain the microphone position information, fetal position information, physiological information of pregnant women that is prone to noise generation, environmental noise information, etc. in the multi-microphone array, so as to more accurately determine the interference targets of different microphones, facilitate subsequent data analysis, and improve the quality of fetal heart sound monitoring;

[0064] A plurality of interference stethoscope weights are divided according to the ratio of the interference stethoscope distance and the interference stethoscope threshold, and whether it is an interference stethoscope target is determined according to the interference stethoscope weight; the interference stethoscope threshold is the minimum interference distance between the noise source and the microphone, and the interference stethoscope threshold is determined based on expert experience and a large amount of data; the source of noise at different distances (for example, but not limited to instrument noise, bowel sounds of pregnant women, stomach peristalsis sounds of pregnant women, etc.) affects the data of fetal heart sound. The main purpose is to reduce unnecessary computing consumption in the actual process, exclude noise that does not affect the quality of fetal heart sound monitoring data, and only screen out targets that have a greater impact on the quality of fetal heart sound monitoring data, reduce the amount of data calculation, improve data screening accuracy, and facilitate subsequent analysis to improve the quality of fetal heart sound monitoring;

[0065] A fetal heart sound stethoscope model analysis module analyzes the first stethoscope data features of the first stethoscope data through the fetal heart sound stethoscope analysis model, and analyzes the second stethoscope data features based on the second stethoscope data; compares the first stethoscope data features with the second stethoscope data features, and performs stethoscope interference processing on the first stethoscope data that has the same features as the second stethoscope data; the first stethoscope data is fetal heart sound monitoring data, and the second stethoscope data is fetal heart sound monitoring interference data; the fetal heart sound monitoring data and the interference data are processed separately to improve the calculation accuracy of the data, prevent errors in the calculation process caused by data mixing, and also improve the accuracy of the fetal heart sound quality;

[0066] Furthermore, the first stethoscope data feature, the second stethoscope data feature and the third stethoscope data feature are obtained by the stethoscope feature data extraction module of the fetal heart sound stethoscope analysis model; the stethoscope feature data extraction module includes a stethoscope data denoising preprocessing unit, a stethoscope data segmentation unit and a stethoscope data feature extraction storage unit; refer to Figure 4 As shown in the stethoscope feature data extraction module; the feature extraction model can extract features from the time series data formed by the monitored fetal heart sound data, facilitating comprehensive data analysis. The deep learning model can understand the comprehensive characteristics of the data, explore the potential features of the monitored fetal heart sound data, improve data recognition efficiency, and facilitate the screening of data interference items, thereby improving data monitoring quality.

[0067] The stethoscope data denoising preprocessing unit performs denoising on the input stethoscope data; preprocesses the denoised input stethoscope data to generate feature-to-be-extracted stethoscope data; the input stethoscope data includes the first stethoscope data, the second stethoscope data and the third stethoscope data;

[0068] The stethoscope data segmentation unit performs data segmentation on the stethoscope data to be extracted with features to generate segmented stethoscope data with features to be extracted;

[0069] Further, a plurality of corresponding second stethoscope data are obtained according to the plurality of interfering stethoscope targets;

[0070] Calculating the average frequency of all the second stethoscope data corresponding to each of the interfering stethoscope targets; calculating the average frequency of all the third stethoscope data;

[0071] Generate a corresponding stethoscope interference label for each of the interfering stethoscope targets and associate it with the corresponding average frequency; segment the second stethoscope data according to the average frequency;

[0072] Obtaining the average frequency of the corresponding stethoscope interference label according to the interference stethoscope weight, and segmenting the first stethoscope data in combination with the average frequency; the segmented first stethoscope data includes multiple groups;

[0073] Since different interference data are different and may be affected by a single interference target or multiple interference targets, the average duration of the interference data is determined by the frequency of the interference data. The interference data and the monitored fetal heart sound data are divided into the same dimension by the average time. While ensuring the recognition accuracy, it can also reduce the calculation amount of model recognition, ensure the accuracy of interference data recognition, facilitate the subsequent elimination of fetal heart sound interference data, and improve data quality;

[0074] The stethoscope data feature extraction unit extracts the stethoscope feature data based on the segmentation feature to be extracted stethoscope data, and performs data association storage; the stethoscope data feature extraction unit includes 2 layers of Transformer layers, 4 layers of LSTM layers, 2 layers of residual convolution layers, 1 layer of feature fusion layer and classification layer; the feature fusion layer adopts tanh function, and the classification layer mainly adopts full connection layer and Softmax function; refer to Figure 5 As shown, the Transformer layer corresponds to Figure 5 The translation model layer in , the LSTM layer corresponds to Figure 5 The memory neural layer in the Transformer layer; the attention mechanism of the Transformer layer can accurately grasp the key data in the data; the LSTM layer is used to extract the time series features of the monitored fetal heart sound data; the residual convolution layer is used to extract the morphological features of the fetal heart sound data. The fusion of time series features and morphological features improves the data dimension, increases the accuracy of data recognition, facilitates subsequent feature similarity comparison calculations, and improves the quality of fetal heart sound monitoring data;

[0075] In this embodiment, four groups of untrained fetal heart rate monitoring data are combined to identify the interference data. The identification results are shown in Table 1:

[0076] Table 14 Recognition accuracy of interference data

[0077] data Interference data recognition accuracy Data Group 01 89.93％ Data Group 02 90.27％ Data Group 03 90.44％ Data Group 04 90.02％

[0078] As can be seen from Table 1, the recognition accuracy of the four groups of data is relatively stable at around 90%, and the recognition accuracy is good;

[0079] Furthermore, the first stethoscope data feature is compared with the second stethoscope data feature, and the first stethoscope data feature is compared with the third stethoscope data feature, and the similarity is calculated by the stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model, with reference to Figure 4 As shown in the stethoscope data feature comparison module, the stethoscope data feature comparison module is:

[0080] σ S =α1·f cos (l in ,l S )+α2·f D (l in ,l S )+α3·f F (l in ,l S )+α4·f B (l in ,l S );

[0081] Among them, σ S is the similarity result, f cos is the cosine similarity, f D is the dynamic time planning similarity, f F is the Fourier similarity, f B is the empirical statistical similarity, l in is the first stethoscope data feature, l S is the data feature being compared, l S Including the second stethoscope data feature and the third stethoscope data feature, α1, α2, α3 and α4 are similarity calculation weights, which default to 0.25 and can be changed according to actual needs; f B It is a statistical result obtained by relevant technical personnel based on a large amount of actual data.

[0082] Based on the identified data features, in order to improve the accuracy of similarity calculation, multi-dimensional features are used for determination, including distance calculation based on cosine vector features, distance calculation based on dynamic time planning algorithm, and Fourier similarity calculation based on time series data frequency and time domain. In order to further improve the accuracy of similarity calculation, statistical methods are added in combination with expert experience for auxiliary calculation. This method can calculate the similarity of data based on different dimensions of data, and can also combine the dual dimensions of algorithm and statistical methods to improve the similarity of data calculation.

[0083] Example 2

[0084] For the sake of specificity, the following content is used to illustrate the model's recognition of abnormal fetal heart sounds:

[0085] a fetal heart sound stethoscope model abnormality module, which obtains third stethoscope data features of third stethoscope data through the fetal heart sound stethoscope analysis model, compares the third stethoscope data features with the first stethoscope data features after stethoscope interference processing, and analyzes abnormal fetal heart sound data, wherein the third stethoscope data is abnormal fetal heart sound data;

[0086] Furthermore, the first stethoscope data feature, the second stethoscope data feature, and the third stethoscope data feature are acquired by a stethoscope feature data extraction module of the fetal heart sound stethoscope analysis model; the stethoscope feature data extraction module includes a stethoscope data noise reduction preprocessing unit, a stethoscope data segmentation unit, and a stethoscope data feature extraction storage unit;

[0087] The stethoscope data denoising preprocessing unit performs denoising on the input stethoscope data; preprocesses the denoised input stethoscope data to generate feature-to-be-extracted stethoscope data; the input stethoscope data includes the first stethoscope data, the second stethoscope data and the third stethoscope data;

[0088] The stethoscope data segmentation unit performs data segmentation on the stethoscope data to be extracted with features to generate segmented stethoscope data with features to be extracted;

[0089] Further, calculating the maximum frequency of all the third stethoscope data; segmenting the third stethoscope data according to the maximum frequency; segmenting the first stethoscope data in combination with the maximum frequency;

[0090] The stethoscope data feature extraction unit extracts the stethoscope feature data based on the segmentation feature to be extracted stethoscope data, and performs data association storage; the stethoscope data feature extraction unit includes 2 layers of Transformer layers, 4 layers of LSTM layers, 2 layers of residual convolution layers, 1 layer of feature fusion layer and a classification layer;

[0091] In this embodiment, the same model was trained using abnormal fetal heart sound data and normal fetal heart sound data. The following shows the highest recognition accuracy of different models during the training process. The number of iterations was set to 150, and an early stopping mechanism was added. The model calculation was stopped after the learning rate reached the same learning rate parameter 5 times. The specific results are shown in Table 2:

[0092] Table 2. Recognition accuracy of abnormal fetal heart sounds under different models

[0093]

[0094] From Table 2, we can see that the Transformer, 4-layer LSTM and 2-layer residual convolution models have higher recognition accuracy than the model in this paper. Figure 6 As shown, the Transformer, 4-layer LSTM and 2-layer residual convolution did not converge on the test set data and showed an upward trend, showing overfitting and poor effect; the loss function of the model in this paper converged after recognition, referring to Figure 7 As shown, the effect is better.

[0095] Furthermore, the first stethoscope data feature and the third stethoscope data feature are compared, and similarity is calculated by a stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model, wherein the stethoscope data feature comparison module is:

[0096] σ S =α1·f cos (l in ,l S )+α2·f D (l in ,l S )+α3·f F (l in ,l S )+α4·f B (l in ,l S );

[0097] Among them, σ S is the similarity result, f cos is the cosine similarity, f D is the dynamic time planning similarity, f F is the Fourier similarity, f B is the empirical statistical similarity, l in is the first stethoscope data feature, l S is the data feature being compared, l S Including the second stethoscope data feature and the third stethoscope data feature, α1, α2, α3 and α4 are similarity calculation weights.

[0098] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time fetal heart sound monitoring system based on a multi-microphone array, characterized in that: include: A fetal heart sound stethoscope array position module is configured to obtain a stethoscope position of a stethoscope target, wherein the stethoscope target includes multiple stethoscope levels and obtain multiple relative stethoscope coordinate centers based on the stethoscope levels; establish a relative stethoscope coordinate system based on the stethoscope position and determine multiple microphone stethoscope positions based on the central stethoscope position of the multi-microphone array; establish a stethoscope-microphone coordinate system based on the central stethoscope position and obtain the microphone stethoscope coordinates of each microphone; the microphone coordinate system is a proportional mapping of the relative stethoscope coordinate system; According to the relative stethoscope coordinate system, the relative stethoscope center coordinates of the central stethoscope position are obtained; according to the relative stethoscope center coordinates and the microphone stethoscope coordinates, the relative stethoscope center coordinates of each microphone are determined, and the relative stethoscope center coordinates of each microphone are the microphone stethoscope position; a fetal heart sound stethoscope data module, which acquires multiple sets of first stethoscope data at different stethoscope positions according to the multi-microphone array; the first stethoscope data is fetal heart sound monitoring data; a fetal heart sound stethoscope interference analysis module, which obtains an interference stethoscope target for each set of the first stethoscope data based on the plurality of sets of the first stethoscope data and the relative stethoscope position coordinates of the microphone stethoscope position; and extracts corresponding second stethoscope data based on the interference stethoscope target; the second stethoscope data being fetal heart sound monitoring interference data; a fetal heart sound stethoscope model analysis module, which analyzes first stethoscope data features of the first stethoscope data using a fetal heart sound stethoscope analysis model, and analyzes second stethoscope data features based on the second stethoscope data; performs a stethoscope data feature comparison between the first stethoscope data features and the second stethoscope data features, and performs stethoscope interference processing on the first stethoscope data having the same features as the second stethoscope data; a fetal heart sound stethoscope model abnormality module, which obtains third stethoscope data features of third stethoscope data through the fetal heart sound stethoscope analysis model, wherein the third stethoscope data is abnormal fetal heart sound data; compares the third stethoscope data features with the first stethoscope data features after stethoscope interference processing, and analyzes the abnormal fetal heart sound data; The first stethoscope data feature, the second stethoscope data feature, and the third stethoscope data feature are acquired by a stethoscope feature data extraction module of the fetal heart sound stethoscope analysis model; the stethoscope feature data extraction module includes a stethoscope data noise reduction preprocessing unit, a stethoscope data segmentation unit, and a stethoscope data feature extraction storage unit; The stethoscope data noise reduction preprocessing unit performs noise reduction processing on the input stethoscope data; Preprocessing is performed based on the noise-reduced input stethoscope data to generate stethoscope data to be extracted with features; the input stethoscope data includes the first stethoscope data, the second stethoscope data, and the third stethoscope data; The stethoscope data segmentation unit performs data segmentation on the feature-to-be-extracted stethoscope data to generate segmented feature-to-be-extracted stethoscope data; and acquires a plurality of corresponding second stethoscope data according to a plurality of the interfering stethoscope targets; Calculating the average frequency of all the second stethoscope data corresponding to each of the interfering stethoscope targets; associating each of the interfering stethoscope targets with a corresponding stethoscope interference label corresponding to the average frequency; and segmenting the second stethoscope data according to the average frequency; Obtaining the average frequency of the corresponding stethoscope interference label according to the interfering stethoscope weight, and segmenting the first stethoscope data in combination with the average frequency; The segmented first stethoscope data includes multiple groups; Calculating the maximum frequency of all the third stethoscope data; segmenting the third stethoscope data according to the maximum frequency; segmenting the first stethoscope data in combination with the maximum frequency; the stethoscope data feature extraction and storage unit extracting the stethoscope feature data based on the segmented feature stethoscope data to be extracted, and performing data association storage; The first stethoscope data feature is compared with the second stethoscope data feature, and the first stethoscope data feature is compared with the third stethoscope data feature. Similarity is calculated by a stethoscope data feature comparison module of the fetal heart sound stethoscope analysis model. The stethoscope data feature comparison module is: ； in, is the similarity result, is the cosine similarity, For dynamic time planning similarity, is the Fourier similarity, is the empirical statistical similarity, is the first stethoscope data feature, is the data feature being compared, including the second stethoscope data feature and the third stethoscope data feature; 、 、 and Calculate weights for similarity.

2. A real-time fetal heart sound monitoring system based on a multi-microphone array according to claim 1, characterized in that: In combination with the relative stethoscope position coordinates of the stethoscope position, obtaining the interference stethoscope target of each group of the first stethoscope data includes: obtaining the relative stethoscope coordinates of each interference target based on the relative stethoscope coordinate center of the stethoscope level; determining the interference stethoscope target of each group of the first stethoscope data based on the interference stethoscope distance of the relative stethoscope center coordinates of each microphone and the relative stethoscope coordinates of each interference target.

3. A real-time fetal heart sound monitoring system based on a multi-microphone array according to claim 2, characterized in that: Determining the interference stethoscope target of each group of the first stethoscope data based on the interference stethoscope distance of each microphone relative to the stethoscope center coordinates and the interference target relative to the stethoscope coordinates also includes: dividing a plurality of interference stethoscope weights based on the ratio of the interference stethoscope distance and the interference stethoscope threshold, and determining whether it is an interference stethoscope target based on the interference stethoscope weight.

4. The real-time fetal heart sound monitoring system based on a multi-microphone array according to claim 1, characterized in that: The stethoscope data feature extraction and storage unit includes 2 Transformer layers, 4 LSTM layers, 2 residual convolution layers, 1 feature fusion layer and a classification layer.

Images