Children home monitoring system based on human body breathing sound and physiological indexes

Abstract

The invention relates to a children home monitoring system based on human body breath sound and physiological indexes, which comprises a client and a server, and is characterized in that the client comprises a data acquisition module used for acquiring a noisy breath sound signal and children body index text data and constructing a reasoning request; the communication module is used for establishing connection with the server side, sending the reasoning request to the server side for preprocessing and prediction, and receiving a dichotomy prediction result which responds to the reasoning request, is obtained by one or more pre-trained models and is used for judging whether the breath sound is abnormal or not from the server side; and the display module is used for displaying the dichotomy prediction result. Compared with the prior art, the breath sound monitoring method has the advantages of realizing breath sound monitoring in children-oriented and home scenes, and being high in accuracy and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of respiratory sound monitoring equipment, and in particular to a child home monitoring system based on human respiratory sounds and physiological indicators. Background Art

[0002] Pneumonia is a common respiratory infection caused by pathogens such as bacteria and viruses, and is especially common in children. Traditional pneumonia diagnosis mainly relies on doctors' clinical experience and examination methods, such as auscultation and X-rays. However, these diagnostic methods require patients to visit medical institutions in person and require a high level of professionalism from doctors. In areas where medical resources are unevenly distributed, it is inconvenient for patients to seek medical treatment, and the demand for telemedicine has increased dramatically, especially during the epidemic.

[0003] With the rapid development of digital technology and artificial intelligence, medical diagnosis has gradually shifted towards intelligence and convenience. In recent years, pneumonia diagnosis methods based on respiratory sound and human physiological sound analysis have begun to attract widespread attention. However, most of these systems are prototype systems in laboratory environments and are still a long way from practical application.

[0004] Although some existing technologies have achieved certain results in respiratory sound analysis and multi-source data fusion, they still have many limitations, especially in the lack of application research in home environments.

[0005] In addition, some existing technologies do not take into account the impact of different stages of children's lung development on monitoring results.

[0006] In summary, there is currently a lack of a children's home breathing sound monitoring system to achieve breathing sound monitoring for children in home scenarios. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a child home monitoring system based on human breathing sounds and physiological indicators, so as to realize breathing sound monitoring for children in home scenarios.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] The present invention provides a child home monitoring system based on human breathing sounds and physiological indicators, including a client and a server, wherein the client includes:

[0010] The data acquisition module is used to obtain noisy breathing sound signals and text data of children's physical indicators and construct inference requests;

[0011] A communication module, used to establish a connection with the server, send the inference request to the server for preprocessing and prediction, and receive from the server a binary classification prediction result of whether the breathing sound is abnormal obtained by one or more pre-trained models in response to the inference request;

[0012] The display module is used to display the binary classification prediction result.

[0013] As a preferred technical solution, the data acquisition module includes:

[0014] The interactive unit is used to display a preset form and obtain the age, gender and subjective feeling information of the child filled in by interacting with the user;

[0015] A body index sensor for collecting real-time body index information of children in text form;

[0016] A text conversion unit, used for combining the age, gender and subjective feeling information with the physical index information to convert physical index text data;

[0017] A request generation unit is used to construct an inference request based on the text data and the identification feature information of the client.

[0018] As a preferred technical solution, the body index sensor includes a body temperature sensor and a heart rate sensor.

[0019] As a preferred technical solution, the server includes:

[0020] An audio preprocessing module, configured to filter the noisy breathing sound signal in the inference request within a non-human voice frequency range and divide the audio into slices of fixed length;

[0021] The text preprocessing module is used to perform outlier removal, standardization and encoding processing on the body indicator text data in the inference request.

[0022] As a preferred technical solution, the server also includes:

[0023] A spectrum feature extraction module is used to extract spectrum features through short-time Fourier transform processing for the respiratory sound signal slice processed by the preprocessing module to obtain Mel-frequency cepstral coefficient features;

[0024] The model warehouse module has a built-in inference model corresponding to multiple ages, which is used to call the inference model matching the inference request according to the identification feature information of the client that issues the inference request to infer the Mel-frequency cepstral coefficient features obtained by the spectrum feature extraction module and the physical indicator text data obtained by the text preprocessing module to obtain a binary classification prediction result of whether one or more corresponding respiratory sounds are abnormal.

[0025] As a preferred technical solution, the types of inference models include logistic regression models and support vector machine models.

[0026] As a preferred technical solution, the model warehouse module includes:

[0027] The model training unit is used to train the inference model according to the set optimization target and training data set, wherein the optimization target of the support vector machine model is:

[0028]

[0029] Among them, ξ i is the slack variable, C is the penalty parameter, N is the number of samples in the training data set, and w is the weight vector of the support vector machine.

[0030] As a preferred technical solution, the display module is also used to display the body indicator text data.

[0031] As a preferred technical solution, the communication module includes a WIFI chip and / or a Bluetooth chip, and the display module includes a display screen.

[0032] As a preferred technical solution, the data acquisition module further includes:

[0033] The breathing sound sensor is used to collect breathing sound signals of children.

[0034] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0035] (1) Realizing respiratory sound monitoring for children in home scenarios: The present invention obtains noisy respiratory sound signals and text data of children's physical indicators through a data acquisition module, and sends them to a server deployed with multiple models through a communication module, obtains a binary classification prediction result of whether the respiratory sound is abnormal, and displays it in an interactive module, thereby realizing respiratory sound monitoring.

[0036] (2) Combining multi-source data for reasoning with high accuracy: The present invention combines multi-source data for respiratory sound monitoring. On the basis of respiratory sound audio, it combines physical indicator text data (including age, gender, body temperature, subjective feelings, etc.). Through multi-source data fusion, it can more comprehensively reflect the patient's health status and improve the accuracy and reliability of diagnosis. In addition, multiple pre-trained models are used to automatically process and analyze the collected respiratory sounds. This process reduces the subjectivity and errors of manual judgment and improves the accuracy of monitoring.

[0037] (3) In view of the characteristics of children, the influence of the degree of lung development on the reasoning results is reduced: the present invention sets up a model warehouse module and maintains reasoning models corresponding to different ages (or age groups) therein. During the reasoning process, the reasoning model matching the child's age is determined according to the request, thereby reducing the influence of the degree of lung development on the reasoning results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of a children's home breathing sound monitoring system based on human physiological sounds and body indicators in an embodiment;

[0039] Figure 2 Schematic diagram of the monitoring process using the children's home respiratory sound monitoring system in the embodiment. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work should fall within the scope of protection of the present invention.

[0041] Regarding the problems existing in the above prior art, see Figure 1 ,This embodiment provides a child home monitoring system based on human breathing sounds and physiological indicators, including a client and a server.

[0042] The client includes:

[0043] (1) Data acquisition module, used to obtain noisy breathing sound signals and children's physical index text data, and construct reasoning requests. Specifically, this module includes:

[0044] The interactive unit is used to display a preset form and obtain the age, gender and subjective feeling information of the child filled in by interacting with the user;

[0045] A body index sensor, used to collect the child's real-time body index information in text form. In this embodiment, the body index sensor includes a body temperature sensor and a heart rate sensor;

[0046] A text conversion unit, used to convert the age, gender, subjective feeling information and physical index information into physical index text data;

[0047] The request generation unit is used to construct an inference request based on the text data and the identification feature information of the client, wherein the identification feature information may be a unique identification number, MAC address, etc. of the user terminal device.

[0048] The breathing sound sensor is used to collect breathing sound signals of children.

[0049] (2) A communication module, including a WIFI chip and / or a Bluetooth chip, for establishing a connection with a server, sending an inference request to the server for preprocessing and prediction, and receiving from the server a binary classification prediction result of whether the respiratory sound is abnormal obtained by one or more pre-trained models in response to the inference request.

[0050] (3) A display module, including a display screen, for displaying binary classification prediction results and / or displaying text data of physical indicators.

[0051] The server includes:

[0052] (1) An audio preprocessing module, which is used to filter the noisy breathing sound signal in the inference request within the non-human voice frequency range and divide the audio into slices of fixed length;

[0053] (2) A text preprocessing module, which is used to remove outliers, standardize, and encode the text data of physical indicators in the inference request.

[0054] (3) a spectrum feature extraction module, which is used to extract spectrum features through short-time Fourier transform processing for the respiratory sound signal slices processed by the preprocessing module to obtain Mel-frequency cepstral coefficient features;

[0055] (4) The model warehouse module has multiple inference models corresponding to different ages (or age groups) built in. It is used to call the inference model that matches the inference request based on the identification feature information of the client that issues the inference request to infer the Mel frequency cepstral coefficient features obtained by the spectrum feature extraction module and the physical index text data obtained by the text preprocessing module to obtain a binary classification prediction result of whether one or more corresponding respiratory sounds are abnormal. By calling different inference models for inference according to the age of the child, the impact of the different degrees of lung development of the child on the inference result is reduced.

[0056] Preferably, with the user's prior consent, the inference request is saved, and when preset conditions are met (for example, the number of saved inference requests reaches a threshold, and / or the inference model is in an idle state), incremental learning is performed on the corresponding inference model.

[0057] Preferably, the model warehouse also has a built-in large language model, which is used to generate suggestion text based on the inference request and the binary classification prediction results, and then send the classification results to the user end, and display them on the display module of the user end.

[0058] In this embodiment, the types of inference models include logistic regression models and support vector machine models.

[0059] The model warehouse module includes a model training unit, which is used to train the inference model according to the set optimization goal and training data set. The optimization goal of the support vector machine model is:

[0060]

[0061] Among them, ξ i is the slack variable, C is the penalty parameter, N is the number of samples in the training data set, and w is the weight vector of the support vector machine.

[0062] See also Figure 2 , the process of using the children's home breathing sound detection breathing sound sensor for monitoring is:

[0063] Step S1, using the breathing sound sensor in the data acquisition module to obtain noisy breathing sound signals (ie, human physiological sounds), and using the body temperature sensor and the heart rate sensor to obtain the child's physical index text data.

[0064] When collecting breathing sound signals, determine the collection environment to ensure that the environmental noise is minimized. The lightweight, easy-to-operate, built-in high-sensitivity breathing sound sensor can accurately capture the subtle sound changes produced during the child's breathing process. The breathing sound sensor has the function of suppressing environmental noise to ensure that high-quality breathing sound signals can still be obtained in a home environment. The portability and ease of use of the breathing sound sensor enable it to be flexibly applied to various scenarios in children's daily lives, such as monitoring during sleep, checking during play, etc., to ensure uninterrupted health monitoring.

[0065] Preferably, the child's body temperature, heart rate and respiratory sound signal-to-noise ratio are collected to determine whether the child is currently in an active or stationary state, and different noise suppression strengths are used according to different states. Further, the microphone array has a directional sound pickup function, and the sound pickup direction is the child's lungs, and the sound pickup direction is pre-set during assembly.

[0066] Specifically, the body temperature sensor and heart rate sensor are used to obtain the text data of children's physical indicators: parents fill in the interactive module in advance and the sensor automatically records, collects the child's age, gender, subjective feelings and other data, and the body temperature sensor and heart rate sensor combine the collected body temperature and heart rate to obtain the body indicator text data. The parent fills in the part using a standardized form to ensure the integrity and consistency of the data, covering basic information such as the child's age, gender, and subjective feelings, such as cough frequency, cough characteristics, etc. The system can automatically record and upload physiological indicators such as body temperature and heart rate to the server. The sensor is usually worn on the wrist or chest and can monitor the health of children 24 hours a day. Body temperature recording sensors such as smart thermometers can measure children's body temperature in real time and alarm in time when abnormal body temperature occurs. In addition, the system can interact with parents through mobile phone applications to provide a visual display of health data, so that parents can grasp the health status of their children in real time. This dual data collection method ensures the comprehensiveness and accuracy of the data and provides a reliable basis for subsequent intelligent diagnosis.

[0067] Step S2, using the communication module to establish a connection with the server, packaging the breathing sound signal and the child's physical index text data into an inference request and sending it to the server for preprocessing and prediction.

[0068] Specifically, the server-side preprocessing and prediction process includes:

[0069] Step 1, audio preprocessing: perform preprocessing such as noise reduction and slicing on the received audio signal to improve the accuracy of subsequent analysis.

[0070] Specifically, a bandpass filter is used to remove non-human voice frequencies, and the filter range refers to the industry standard. The median filter method is used for noise reduction, and the window size is set to an appropriate value. The audio signal is divided into slices of fixed length.

[0071] Step 2, audio feature extraction: Use the audio classification model to extract features from the processed audio, identify and quantify audio features such as moist rales and wheezing.

[0072] Specifically, short-time Fourier transform (STFT) is used to extract spectral features and convert the time domain signal into time-frequency representation. The principle is:

[0073]

[0074] Mel-frequency cepstral coefficient (MFCC) features are calculated, the spectrum features are further processed, and the first N coefficients are selected as features used in subsequent tasks, making them more suitable for machine learning and pattern recognition tasks.

[0075] Step 3, text cleaning: remove irrelevant information and noise to ensure the accuracy and consistency of the data.

[0076] Specifically, delete missing data and outliers to standardize numerical data, such as the standardization formula for body temperature:

[0077]

[0078] where μ T and σ T are the mean and standard deviation of body temperature, respectively.

[0079] Step 4, feature engineering: extract key physical indicators such as age, gender, and body temperature as text features.

[0080] Specifically, age is converted into discrete features, such as infant, toddler, preschooler, etc. Gender features are processed using one-hot encoding. The standardized body temperature change is calculated: ΔTbody.

[0081] Step 5, feature fusion: The audio features obtained by audio classification are fused with the text features of physical indicators to form a comprehensive feature vector, which serves as the input of subsequent multiple models.

[0082] Specifically, the audio feature vector Faudio and the text feature vector Ftext are concatenated to form a comprehensive feature vector Fcombined = [Faudio, Ftext] as the input of subsequent multiple models.

[0083] Specifically, the model training includes the following steps:

[0084] Step 1, model selection: Use multiple machine learning models such as logistic regression, support vector machine, adaptive boosting, etc. to conduct binary classification experiments to determine whether there is an abnormality.

[0085] Specifically, a logistic regression model is used, and the model formula is:

[0086]

[0087] Using support vector machine (SVM), the optimization objective is:

[0088]

[0089] where ξ iis a slack variable, C is a penalty parameter, N is the number of samples in the training data set, that is, N samples are involved in the training, and w is the weight vector of the support vector machine, which is a parameter that the model needs to learn and is used to define the decision boundary. Slack variables are used to deal with inseparable samples in the data set, indicating the degree of violation of the decision boundary by sample i. C is a penalty parameter that is used to weigh the complexity of the decision boundary and the number of misclassified samples. A larger C value can make the model pay more attention to the correct classification (i.e., reduce misclassification), while a smaller C value can make the model pay more attention to the smoothness of the decision boundary.

[0090] Preferably, a classification model is also included to determine whether the respiratory sounds include moist rales, wheezing sounds, etc.

[0091] Step 2, model training and optimization: Use labeled data sets to train the model, and improve model performance through cross-validation, parameter tuning, and other means.

[0092] Specifically, K-fold cross validation is used, with K = 5. Model hyperparameters are adjusted through grid search.

[0093] Step 3, application of large language model: introduce the large language model and compare its diagnostic effect with that of the traditional machine learning model to further improve the judgment accuracy.

[0094] Specifically, a large language model, such as GPT4 or BERT, is used to process text features and audio features, and give home monitoring results.

[0095] Step 4, model evaluation: Evaluate model performance through indicators such as accuracy, precision, recall, F1 score, etc. Use confusion matrix to analyze model classification effect and identify misclassification.

[0096] Specific:

[0097] Calculate the accuracy:

[0098] Calculate the precision:

[0099] Calculate the recall rate (Recall):

[0100] Calculate the F1 score:

[0101] Construct a confusion matrix with elements TP, TN, FP, and FN. Calculate the accuracy, precision, and recall of each category. TP (True Positive): The number of samples whose true category is positive and correctly predicted as positive. TN (True Negative): The number of samples whose true category is negative and correctly predicted as negative. FP (False Positive): The number of samples whose true category is negative but incorrectly predicted as positive. FN (False Negative): The number of samples whose true category is positive but incorrectly predicted as negative.

[0102] Step 5, model validation: implement cross-validation to ensure the stability and generalization ability of the model on different data sets.

[0103] Specifically, multiple cross-validation experiments were performed, and the validation results were recorded each time. The mean and variance of the cross-validation were calculated to evaluate the stability of the model.

[0104] Step S3, obtaining the binary classification prediction results of whether the breathing sound is abnormal from multiple models pre-trained in the server, and displaying them in the interactive module.

[0105] The content finally displayed in the interactive module includes but is not limited to the binary classification results of multiple models (i.e. whether it is abnormal), physical indicator information (including body temperature, heart rate, etc.), and home monitoring results generated by the large language model (including health status analysis and suggestion information).

[0106] Preferably, the interactive module also displays the respiratory sound classification results obtained by the classification model.

[0107] This embodiment is aimed at the scenario of children's breathing sound monitoring at home. It is based on breathing sound audio analysis and physical index text processing, combined with machine learning algorithms, collects data through portable breathing sound sensors, and uses the network to achieve remote monitoring. Efficient and accurate monitoring of children's breathing sounds at home is achieved. Breathing sounds contain rich pathological information, such as moist rales, wheezing, etc., which can assist in health judgment through audio signal processing and feature extraction technology. In addition, this realizes the automatic classification of breathing sounds, which significantly improves the accuracy of diagnosis.

[0108] By fusing respiratory sound audio and body index text data (such as age, gender, body temperature, cough symptoms, etc.) through multi-source data, the patient's health status can be more comprehensively reflected, and the accuracy and reliability of diagnosis can be improved. After feature extraction, the audio data is standardized with text data (such as age, gender, body temperature, etc.), and fused into a unified feature vector, which is input into the machine learning model for classification and judgment. In this way, not only can the pathological information in the audio signal be utilized, but also the personal health information of the child can be combined to improve the overall accuracy of the judgment. Through portable audio acquisition and smart wearable devices, children can collect respiratory sound and body index data at home, and transmit them to the server through the Internet for analysis and judgment. This model not only reduces the number of times patients go to the hospital, but also provides a new way for the rational allocation of medical resources.

[0109] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A child home monitoring system based on human breathing sounds and physiological indicators, characterized in that: It includes a client and a server, wherein the client includes: The data acquisition module is used to obtain noisy breathing sound signals and text data of children's physical indicators and construct inference requests; A communication module, used to establish a connection with the server, send the inference request to the server for preprocessing and prediction, and receive from the server a binary classification prediction result of whether the breathing sound is abnormal obtained by one or more pre-trained models in response to the inference request; The display module is used to display the binary classification prediction result.

2. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The data acquisition module comprises: The interactive unit is used to display a preset form and obtain the age, gender and subjective feeling information of the child filled in by interacting with the user; A body index sensor for collecting real-time text-based body index information of children; A text conversion unit, used for combining the age, gender and subjective feeling information with the physical index information to convert physical index text data; A request generating unit is used to construct an inference request based on the text data and the identification feature information of the client.

3. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The body index sensor includes a body temperature sensor and a heart rate sensor.

4. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The server includes: An audio preprocessing module, configured to filter the noisy breathing sound signal in the inference request within a non-human voice frequency range and divide the audio into slices of fixed length; The text preprocessing module is used to perform outlier removal, standardization and encoding processing on the body indicator text data in the inference request.

5. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 4, characterized in that: The server also includes: A spectrum feature extraction module is used to extract spectrum features through short-time Fourier transform processing for the respiratory sound signal slice processed by the preprocessing module to obtain Mel-frequency cepstral coefficient features; The model warehouse module has a built-in inference model corresponding to multiple ages, which is used to call the inference model matching the inference request according to the identification feature information of the client that issues the inference request to infer the Mel-frequency cepstral coefficient features obtained by the spectrum feature extraction module and the physical indicator text data obtained by the text preprocessing module to obtain a binary classification prediction result of whether one or more corresponding respiratory sounds are abnormal.

6. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 5, characterized in that: The types of inference models include logistic regression models and support vector machine models.

7. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 6, characterized in that: The model warehouse module includes: The model training unit is used to train the inference model according to the set optimization target and training data set, wherein the optimization target of the support vector machine model is: Among them, ξ i is the slack variable, C is the penalty parameter, N is the number of samples in the training data set, and w is the weight vector of the support vector machine.

8. The child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The display module is also used to display the body index text data.

9. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The communication module includes a WIFI chip and / or a Bluetooth chip, and the display module includes a display screen.

10. A child home monitoring system based on human breathing sounds and physiological indicators according to claim 1, characterized in that: The data acquisition module also includes: The breathing sound sensor is used to collect breathing sound signals of children.

Images