CN115444367A

CN115444367A - Monitoring method and system of respiratory function rehabilitation training instrument based on artificial intelligence

Info

Publication number: CN115444367A
Application number: CN202211062173.7A
Authority: CN
Inventors: 高天; 王倩; 林仙枝
Original assignee: Guangxi Rio Tinto Medical Technology Co ltd
Current assignee: Guangxi Rio Tinto Medical Technology Co ltd
Priority date: 2022-09-01
Filing date: 2022-09-01
Publication date: 2022-12-09
Anticipated expiration: 2042-09-01
Also published as: CN115444367B

Abstract

The invention discloses a monitoring method and a system of a respiratory function rehabilitation training instrument based on artificial intelligence, which comprises the following steps: responding to the respiratory function rehabilitation monitoring request by using a server, and calling a first respiratory state identification model for processing to obtain respiratory state parameters after respiratory physiological parameters are converted; and a plurality of target users with continuous abnormal breathing state characteristics are determined based on the breathing state parameters, and corresponding medical instrument product push information and rehabilitation training information push information are pushed for each target user.

Description

Monitoring method and system of respiratory function rehabilitation training instrument based on artificial intelligence

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a monitoring method and a monitoring system of a respiratory function rehabilitation training instrument based on artificial intelligence.

Background

At present, a corresponding rehabilitation instrument is generally configured for the respiratory function rehabilitation training of a patient, the physiological parameters of the corresponding patient can be recorded besides the auxiliary training, and because the condition of each patient is different, the adopted rehabilitation training strategy is different, and the corresponding acquired physiological parameters are different, the judgment of the pathological condition of the patient determined by the respiratory function rehabilitation training instrument is inconvenient, and the difficulty is caused for the follow-up push of the targeted recommendation of medical equipment products and the rehabilitation training consultation.

Disclosure of Invention

The invention aims to provide a monitoring method and a monitoring system of a respiratory function rehabilitation training instrument based on artificial intelligence.

In a first aspect, an embodiment of the present invention provides a monitoring system for a respiratory function rehabilitation training instrument based on artificial intelligence, where the monitoring system for a respiratory function rehabilitation training instrument based on artificial intelligence includes a server and a respiratory function rehabilitation training instrument in communication connection with the server;

the respiratory function rehabilitation training instrument is used for calling the configured sensor to acquire the physiological parameters of the user;

the server is used for responding to the respiratory function rehabilitation monitoring request and establishing communication connection with the respiratory function rehabilitation training instrument; acquiring physiological parameters of a user from a respiratory function rehabilitation training instrument; acquiring a respiratory physiological parameter to be evaluated from a user physiological parameter, and calling a first respiratory state identification model for processing to obtain a respiratory state parameter after the respiratory physiological parameter is converted; evaluating according to the breathing state parameters by combining a preset breathing function rehabilitation training table to obtain a monitoring result of the breathing function rehabilitation training instrument; determining a plurality of target users with continuous abnormal breathing state characteristics based on the monitoring result of the breathing function training instrument; and determining target on-line pushing information matched with each target user according to the continuous abnormal breathing state characteristics of each target user, wherein the target on-line pushing information comprises pushing information of medical equipment products and pushing information of rehabilitation training information.

In a second aspect, an embodiment of the present invention provides a monitoring method for a respiratory function rehabilitation training instrument based on artificial intelligence, including:

responding to the respiratory function rehabilitation monitoring request, and establishing communication connection with a respiratory function rehabilitation training instrument;

acquiring a user physiological parameter from a respiratory function rehabilitation training instrument, wherein the user physiological parameter is acquired by a sensor configured for the respiratory function rehabilitation training instrument;

acquiring a respiratory physiological parameter to be evaluated from a user physiological parameter, and calling a first respiratory state identification model for processing to obtain a respiratory state parameter after the respiratory physiological parameter is converted;

evaluating according to the respiratory state parameters by combining a preset respiratory function rehabilitation training table to obtain a monitoring result of the respiratory function rehabilitation training instrument;

obtaining a monitoring result of the respiratory function training instrument;

determining a plurality of target users with continuous abnormal breathing state characteristics based on the monitoring result of the breathing function training instrument;

and determining target on-line pushing information matched with each target user according to the continuous abnormal breathing state characteristics of each target user, wherein the target on-line pushing information comprises pushing information of medical equipment products and pushing information of rehabilitation training information.

Compared with the prior art, the invention has the beneficial effects that: the invention discloses a monitoring system of a respiratory function rehabilitation training instrument based on artificial intelligence, which comprises: responding to the respiratory function rehabilitation monitoring request by using a server, and calling a first respiratory state recognition model for processing to obtain respiratory state parameters after respiratory physiological parameter conversion; and a plurality of target users with continuous abnormal breathing state characteristics are determined based on the breathing state parameters, and corresponding medical instrument product push information and rehabilitation training information push information are pushed for each target user.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly described below. It is appreciated that the following drawings depict only certain embodiments of the invention and are therefore not to be considered limiting of its scope. For a person skilled in the art, it is possible to derive other relevant figures from these figures without inventive effort.

Fig. 1 is a schematic flow chart illustrating steps of a monitoring method of an artificial intelligence based respiratory function rehabilitation training device according to an embodiment of the present invention;

fig. 2 is a schematic block diagram of a monitoring system of a respiratory function rehabilitation training instrument based on artificial intelligence according to an embodiment of the present invention;

fig. 3 is a schematic block diagram of a structure of a computer device according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention.

Referring to fig. 1, fig. 1 is a schematic diagram illustrating a process flow of a monitoring method of a respiratory function rehabilitation training apparatus based on artificial intelligence according to an embodiment of the present invention, where the monitoring method of the respiratory function rehabilitation training apparatus based on artificial intelligence can be implemented by using a server in a monitoring system of the respiratory function rehabilitation training apparatus based on artificial intelligence as an execution main body, and the server is in communication connection with the respiratory function rehabilitation training apparatus, and a specific process flow of the monitoring method of the respiratory function rehabilitation training apparatus based on artificial intelligence may be as follows:

210. and responding to the respiratory function rehabilitation monitoring request, and establishing communication connection with the respiratory function rehabilitation training instrument.

220. And acquiring the physiological parameters of the user from the respiratory function rehabilitation training instrument, wherein the physiological parameters of the user are acquired by a sensor configured for the respiratory function rehabilitation training instrument.

230. And acquiring the respiratory physiological parameters to be evaluated from the physiological parameters of the user, and calling the first respiratory state identification model for processing to obtain the respiratory state parameters after the respiratory physiological parameters are converted.

240. And evaluating according to the breathing state parameters by combining a preset breathing function rehabilitation training table to obtain a monitoring result of the breathing function rehabilitation training instrument.

250. And obtaining a monitoring result of the respiratory function training instrument.

260. And determining a plurality of target users with continuous abnormal breathing state characteristics based on the monitoring result of the respiratory function training instrument.

270. Determining target on-line pushing information matched with each target user according to the continuous abnormal breathing state characteristics of each target user, wherein the target on-line pushing information comprises pushing information of medical equipment products and pushing information of rehabilitation training information.

In the embodiment of the invention, the respiratory function rehabilitation training instrument can comprise a plurality of sensing devices for collecting a plurality of sensing data of a human body in real time, for example, aiming at a respiratory rehabilitation scene, the corresponding sensors can be called to monitor indexes such as electrocardio, oxyhemoglobin saturation, blood pressure, respiratory frequency and the like, so that when the respiratory function rehabilitation monitoring is required, the respiratory function rehabilitation training instrument can be firstly in communication connection with the respiratory function rehabilitation training instrument to achieve the purpose of acquiring accurate parameters in real time. After the physiological parameters of the user are obtained, the first respiratory state recognition model can be called to process the physiological parameters to obtain the respiratory state parameters after the respiratory physiological parameters are converted, it should be understood that the physiological parameters of the user are parameters which intuitively reflect the body state of the patient, the number of indexes is large, the physiological parameters are objective data, in order to quickly and accurately recognize the respiratory state of the user, the physiological data need to be converted into the respiratory state parameters (such as a corresponding identifier or a corresponding numerical value, which is not limited herein) which can intuitively reflect the current specific situation, then the monitoring results of the respiratory function rehabilitation training instrument can be obtained, according to the monitoring results of the respiratory function rehabilitation training instrument, a plurality of target users with continuous abnormal respiratory state characteristics can be determined, and according to the monitoring results corresponding to each user, the push information of each user comprises push information of medical instrument products and push information of rehabilitation training information push information, so that the counseling required by the rehabilitation of the user can be provided for the user without accuracy, rather than the targeted unified counseling push scheme in the prior art.

In order to more clearly describe the scheme provided by the embodiment of the present application, the step 230 may be implemented by the following detailed steps.

101. Acquiring the respiratory physiological parameters to be evaluated, wherein the respiratory physiological parameters comprise at least one respiratory physiological index.

The physiological parameters of respiration are analyzed to obtain various physiological indexes of respiration in the physiological parameters of respiration, and in the embodiment of the invention, the physiological indexes of respiration include, but are not limited to, indexes of respiration frequency, blood sugar, heartbeat, blood oxygen concentration and the like. Alternatively, in some embodiments, the respiratory physiological parameter may be converted from corresponding sensor data.

102. The method comprises the steps of obtaining a respiratory characteristic vector corresponding to at least one respiratory physiological index in respiratory physiological parameters through a first respiratory state recognition model, wherein the first respiratory state recognition model is obtained through training according to a plurality of sample respiratory data pairs, each sample respiratory data pair comprises sample respiratory state data and sample respiratory physiological data obtained through converting the sample respiratory state data through a second respiratory state recognition model, the second respiratory state recognition model is obtained through training according to optimized respiratory data pairs, one optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data, and the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data.

The types of the first respiratory state identification model and the second respiratory state identification model may be Convolutional Neural Network (CNN), recurrent Neural Network (RNN), long Short Term Memory (LSTM), bidirectional Recurrent Neural Network (BiRNN), and the like. It should be noted that the above examples should not be construed as limiting the first and second respiratory state identification models.

The second respiratory state recognition model is obtained by training according to the optimized respiratory data pair, one optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of corresponding optimized sample respiratory physiological data with the same scene, and the second respiratory state recognition model is obtained by training according to a plurality of mapping data. The trained second respiratory state recognition model converts the sample respiratory state data, and can output more diversified respiratory physiological side corresponding user states.

103. And predicting at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes according to the respiratory characteristic vector by using the first respiratory state recognition model, and converting the respiratory physiological indexes into first confidence degrees of the corresponding pre-stored respiratory state indexes.

Optionally, in some embodiments, the step of "predicting, by using the first respiratory state identification model, at least two pre-stored respiratory state indicators corresponding to each respiratory physiological indicator according to the respiratory feature vector, and converting each respiratory physiological indicator into a first confidence of the corresponding pre-stored respiratory state indicator" may include:

the breathing characteristic vector is corresponding to a second breathing characteristic matrix through the matching relation between a first breathing characteristic matrix at the breathing physiology side and a second breathing characteristic matrix at the breathing state side in the first breathing state recognition model, and a reference breathing characteristic vector is obtained;

and determining at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes in the respiratory physiological index set according to the matching degree of the respiratory characteristic vectors of the respiratory physiological index set at the respiratory state side in the second respiratory characteristic matrix and the reference respiratory characteristic vector, and converting the respiratory physiological indexes into first confidence coefficients of the corresponding pre-stored respiratory state indexes.

The respiratory physiological index set belongs to the respiratory state side, and can be composed of a plurality of respiratory physiological indexes.

The reference respiratory feature vector is obtained by corresponding the respiratory feature vector corresponding to the respiratory physiological index to the second respiratory feature matrix, and specifically, convolution operation and pooling operation can be performed on the respiratory feature vector of the respiratory physiological index to obtain the reference respiratory feature vector corresponding to the second respiratory feature matrix.

The step "determining at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes in the respiratory physiological index set according to the respiratory characteristic vectors of the respiratory physiological index set on the respiratory state side in the second respiratory characteristic matrix and the matching degree of the reference respiratory characteristic vectors, and converting each respiratory physiological index into a first confidence coefficient of a corresponding pre-stored respiratory state index", may include:

calculating the matching degree of the respiratory characteristic vector of the respiratory physiological index set at the respiratory state side in the second respiratory characteristic matrix and the reference respiratory characteristic vector;

determining at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes in the respiratory physiological index set according to the matching degree; and predicting a first confidence coefficient of each respiratory physiological index converted into a corresponding prestored respiratory state index.

The matching degree may be a euclidean distance, a cosine distance, a manhattan distance, and the like, which is not limited in this embodiment. The smaller the matching degree is, the larger the difference between the respiratory physiological index set and the scene of the corresponding respiratory physiological index is, and the larger the matching degree is, the closer the respiratory physiological index set and the scene of the corresponding respiratory physiological index are.

In this embodiment, the respiration physiological index set with the matching degree smaller than the preset distance may be determined as the pre-stored respiration state index, where the preset distance may be set according to an actual situation, and this embodiment does not limit this. For example, the number of pre-stored breathing state indicators that need to be obtained may be set according to the number of the pre-stored breathing state indicators.

And predicting a first confidence coefficient of each respiratory physiological index converted into a corresponding prestored respiratory state index through a full connection layer in the first respiratory state identification model.

104. And determining a predicted respiratory state index corresponding to each respiratory physiological index from the pre-stored respiratory state indexes according to the first confidence coefficient through the first respiratory state identification model.

For each respiratory physiological index, the pre-stored respiratory state index with the maximum first confidence coefficient can be used as the predicted respiratory state index corresponding to the respiratory physiological index.

105. And performing weighted average operation on each predicted respiratory state index through the first respiratory state identification model to obtain the respiratory state parameter after the respiratory physiological parameter is converted.

The first respiratory state recognition model and the second respiratory state recognition model are trained, and the second respiratory state recognition model is obtained by training according to a plurality of mapping data, so that sample respiratory physiological data in a sample respiratory data pair constructed by the second respiratory state recognition model are relatively more diverse; the first respiratory state recognition model is obtained by training sample respiratory data constructed according to the second respiratory state recognition model, and the diversity of the respiratory state parameters of a preset user in the sample respiratory data pair can be increased for the first respiratory state recognition model, so that the conversion quality of the first respiratory state recognition model is enhanced through richer information.

The fusion mode of the predicted respiratory state indexes may specifically be that all the predicted respiratory state indexes are spliced according to a certain mode to obtain respiratory state parameters.

It should be noted that the first respiratory state identification model may be trained from a plurality of sample training sets, and the sample training sets may include sample respiratory data pairs and initial respiratory data pairs, where each initial respiratory data pair includes paired initial sample respiratory physiological data and initial sample respiratory state data, and the initial sample respiratory physiological data and the initial sample respiratory state data have the same scenario. The first respiratory state recognition model may be provided to the conversion device based on artificial intelligence after being trained by another device, or may be trained by the conversion device based on artificial intelligence.

If the conversion device according to artificial intelligence is used for training, before the step "acquiring the physiological parameter of respiration to be evaluated", the following example is further provided in the embodiment of the present invention:

acquiring sample respiratory state data and a plurality of initial respiratory data pairs, wherein each initial respiratory data pair comprises paired initial sample respiratory physiological data and initial sample respiratory state data;

converting the sample respiratory state data through a second respiratory state identification model to obtain converted sample respiratory physiological data, and combining the converted sample respiratory physiological data with corresponding sample respiratory state data to form a sample respiratory data pair;

determining a sample respiration data pair and an initial respiration data pair as a sample training set of a first respiration state identification model, wherein the initial sample respiration physiological data and the sample respiration physiological data are original sample respiration physiological data, and the initial sample respiration state data and the sample respiration state data are original sample respiration state data;

the method comprises the steps of converting original sample respiratory physiological data through a first respiratory state recognition model to obtain converted predicted sample respiratory state data, and performing conversion training from a respiratory physiological side to a respiratory state side on the first respiratory state recognition model according to a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data.

Wherein, a group of sample training sets comprises pairs of original sample respiratory physiological data and original sample respiratory state data, and the sample training sets can be considered to be obtained by combining the sample respiratory data pairs and the initial respiratory data pairs.

The source data of the first respiratory state identification model is the target data of the second respiratory state identification model, and the target data of the first respiratory state identification model is the source data of the second respiratory state identification model, that is, the source data of the first respiratory state identification model is the respiratory physiology side, and the target data of the first respiratory state identification model is the respiratory state side. It is to be understood that the first respiratory state identification model may be regarded as a forward respiratory state identification model and the second respiratory state identification model as a reverse respiratory state identification model. Where a forward direction may be considered a source-to-target data direction and a reverse direction may be considered a target-to-source data direction.

The sample respiratory state data can be converted into sample respiratory physiological data (namely, the sample respiratory physiological data mentioned in the previous embodiment) through the reverse respiratory state identification model to construct a sample respiratory data pair, and the sample respiratory state data pair is merged with the optimized sample respiratory state data set to enhance data, so that the quality of the forward respiratory state identification model is improved, and the following examples can be seen in the process:

1001. training a reverse breathing state recognition model by using a diversity-driven training target, wherein the diversity specifically refers to the diversity of mapping data; specifically, optimized sample respiratory state data set pairs can be obtained, each optimized sample respiratory state data set pair can comprise optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data corresponding to the optimized sample respiratory state data, the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data, the optimized sample respiratory state data are converted through a reverse respiratory state recognition model to obtain converted predicted sample respiratory physiological data, and the reverse respiratory state recognition model is trained according to cost functions between the mapping data corresponding to the optimized sample respiratory state data and the predicted sample respiratory physiological data;

1002. converting sample respiratory state data by using a trained reverse respiratory state identification model, and outputting more diversified identification data, wherein the identification data is sample respiratory physiological data, and the sample respiratory physiological data and the corresponding sample respiratory state data form a sample respiratory data pair, so that the information of the constructed sample respiratory data pair on source data is enriched;

1003. combining the sample breath data pair with the optimized sample breath state data set pair to obtain combined sample training data, wherein the combined sample training data comprises combined optimized sample breath physiological data and combined optimized sample breath state data; and training the forward respiratory state recognition model according to the cross entropy of the maximum likelihood estimation between the respiratory state data of the prediction sample and the respiratory state data of the corresponding optimization sample.

In one embodiment, if the respiratory physiological side is chinese and the respiratory state side is english, the first respiratory state recognition model is a respiratory state recognition model in chinese translation, and the second respiratory state recognition model is a respiratory state recognition model in english translation.

It is emphasized, among other things, that the second respiratory state recognition model is already trained. Because the second respiratory state identification model is obtained by training according to a plurality of mapping data, the sample respiratory physiological data in the sample respiratory data pair constructed by the second respiratory state identification model is relatively more diverse. For the first respiratory state identification model, the diversity of the respiratory state parameters of the preset user in the sample respiratory data pair can be increased, and the preset respiratory state data are enriched.

Optionally, in some embodiments, the step "converting the original sample respiratory physiological data through the first respiratory state identification model to obtain converted predicted sample respiratory state data, and performing conversion training from a respiratory physiological side to a respiratory state side on the first respiratory state identification model according to a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data" may include:

acquiring a respiration characteristic vector corresponding to each sample respiration physiological index in the original sample respiration physiological data through a first respiration state identification model;

predicting at least two undetermined sample respiratory physiological indexes corresponding to each sample respiratory physiological index according to the respiratory feature vector, and converting each sample respiratory physiological index into a fourth confidence coefficient of the corresponding undetermined sample respiratory physiological index;

according to the fourth confidence coefficient, determining target indexes corresponding to the breathing physiological indexes of the samples from the breathing physiological indexes of the undetermined samples, and performing weighted average operation on the target indexes to obtain the breathing state data of the predicted samples;

calculating a first cost function between the predicted sample respiratory state data and original sample respiratory state data corresponding to the original sample respiratory physiological data;

and optimizing the model architecture of the first respiratory state recognition model according to the first cost function to obtain the trained first respiratory state recognition model.

In the optimization process of the first respiratory state identification model, a fourth confidence coefficient of a respiratory physiological index of an undetermined sample, which is close to the application scene matching degree in the mapping data (namely, the corresponding original sample respiratory state data), is increased.

The above embodiment may be referred to in the specific process of predicting the respiratory characteristic vector of the to-be-determined sample corresponding to each sample respiratory physiological index according to the respiratory characteristic vector of each sample respiratory physiological index. The step of predicting at least two undetermined sample respiratory physiological indexes corresponding to each sample respiratory physiological index according to the respiratory feature vector and converting each sample respiratory physiological index into a fourth confidence coefficient of the corresponding undetermined sample respiratory physiological index may include:

and determining at least two undetermined sample respiratory physiological indexes corresponding to each sample respiratory physiological index in the respiratory physiological index set according to the matching degree of the respiratory characteristic vector of the respiratory physiological index set at the respiratory state side in the second respiratory characteristic matrix and the reference respiratory characteristic vector, and converting each sample respiratory physiological index into a fourth confidence coefficient of the corresponding undetermined sample respiratory physiological index.

For each sample respiratory physiological index, the undetermined sample respiratory physiological index with the maximum fourth confidence coefficient can be used as a target index corresponding to the sample respiratory physiological index.

When the first respiratory state identification model outputs each data in the sample respiratory state data, the first respiratory state identification model is not only guided by the mapping data (namely, the corresponding original sample respiratory state data) in the training set, but also considers the state data which is not covered by the reference data in the mapping data but can be more accurate.

The step of calculating a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data may include:

calculating index difference cost of each target index in the predicted sample respiratory state data and corresponding original data in the original sample respiratory state data;

and performing weighted average operation on the index difference cost corresponding to each target index in the respiratory state data of the prediction sample to obtain a first cost function between the respiratory state data of the prediction sample and the respiratory state data of the original sample corresponding to the respiratory physiological data of the original sample.

The index difference cost can be calculated according to the breathing characteristic vector corresponding to the target index and the vector distance between the breathing characteristic vectors corresponding to the original data. The vector distance may include a euclidean distance or a cosine distance, etc.

The fusion of the index difference costs corresponding to the target indexes may specifically be to perform weighted summation on the index difference costs corresponding to the target indexes to obtain a first cost function.

Optionally, in some embodiments, the step of "calculating a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data" may include:

obtaining the confidence coefficient of original data in the original sample respiratory state data converted from each sample respiratory physiological index;

and calculating a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data according to the confidence coefficient.

Optionally, in some embodiments, the step "optimizing a model architecture of the first respiratory state recognition model according to the first cost function to obtain the trained first respiratory state recognition model" may specifically include: and optimizing the model architecture of the first respiratory state recognition model by adopting a back propagation algorithm (BP) or a Stochastic Gradient Descent (SGD), and optimizing the model architecture of the first respiratory state recognition model according to the first cost function so that the first cost function is smaller than a preset cost function to obtain the trained first respiratory state recognition model. The preset cost function can be set according to actual conditions.

The second respiratory state identification model is obtained by training according to the optimized respiratory data pair, and one optimized respiratory data pair comprises one optimized sample respiratory state data and a plurality of corresponding optimized sample respiratory physiological data. The second breathing state recognition model may be provided to the conversion device based on artificial intelligence after being trained by another device, or may be trained by the conversion device based on artificial intelligence.

Optionally, in this embodiment, the first respiratory state identification model may be obtained by training in advance according to the assisted respiration data pair. A pair of assisted breathing data includes a target sample breathing physiology data and a corresponding plurality of optimized sample breathing state data, the optimized sample breathing state data being a reference transformation of the target sample breathing physiology data. The first respiratory state recognition model may be provided to the conversion device based on artificial intelligence after being trained by another device, or may be trained by the conversion device based on artificial intelligence.

(1) And training a second respiratory state recognition model.

If the second respiratory state recognition model is trained by the artificial intelligence-based conversion device, before the step "converting the sample respiratory state data through the second respiratory state recognition model to obtain the converted sample respiratory physiological data", the following example is further provided:

acquiring optimized respiration data pairs, wherein one optimized respiration data pair comprises optimized sample respiration state data and a plurality of corresponding optimized sample respiration physiological data, and the optimized sample respiration physiological data is reference conversion of the optimized sample respiration state data;

converting the optimized sample respiratory state data through a preset second respiratory state identification model to obtain converted predicted sample respiratory physiological data;

calculating a second cost function between the predicted sample respiratory physiological data of the same optimized sample respiratory state data and the corresponding optimized sample respiratory physiological data;

and optimizing the model architecture of the preset second respiratory state identification model according to the second cost function to obtain the second respiratory state identification model.

The model architecture of the preset second respiratory state identification model can be optimized by adopting a back propagation algorithm or a random gradient descent algorithm, and the model architecture of the preset second respiratory state identification model is optimized according to the second cost function, so that the second cost function is smaller than the preset cost function, and the second respiratory state identification model is obtained. The preset cost function can be set according to actual conditions.

Optionally, in some embodiments, the step of "converting the optimized sample respiratory state data by presetting the second respiratory state identification model to obtain the converted predicted sample respiratory physiological data" may include:

acquiring a respiratory feature vector corresponding to each target index in the optimized sample respiratory state data through a preset second respiratory state identification model;

predicting at least two pre-stored respiratory physiological index variables corresponding to each target index according to the respiratory feature vector, and converting each target index into a second confidence coefficient of the corresponding pre-stored respiratory physiological index variable;

according to the second confidence coefficient, determining target respiratory physiological index variables corresponding to each target index from prestored respiratory physiological index variables;

and performing weighted average operation on each target respiration physiological index variable to obtain predicted sample respiration physiological data after optimized sample respiration state data conversion.

For each target index, the prestored breathing physiological index variable with the largest second confidence coefficient can be used as the target breathing physiological index variable corresponding to the target index, and the target breathing physiological index variable can also be regarded as the conversion result of the target index. The fusion mode of the target respiration physiological index variables may specifically be that the target respiration physiological index variables are spliced according to a certain mode to obtain the prediction sample respiration physiological data.

Optionally, in some embodiments, the step of "predicting at least two pre-stored physiological indicator variables corresponding to each target indicator according to the respiratory feature vector, and converting each target indicator into a second confidence of the corresponding pre-stored physiological indicator variable" may include:

acquiring a respiratory physiological index variable set, wherein the respiratory physiological index variable set comprises a plurality of respiratory physiological index variables;

corresponding the respiratory characteristic vector of the target index to the first respiratory characteristic matrix through the matching relation between a second respiratory characteristic matrix at the respiratory state side and a first respiratory characteristic matrix at the respiratory physiology side in a preset second respiratory state recognition model to obtain a first reference respiratory characteristic vector;

and determining at least two pre-stored respiratory physiological index variables corresponding to each target index in the respiratory physiological index variables and a second confidence coefficient of converting each target index into the corresponding pre-stored respiratory physiological index variable according to the matching degree of the respiratory characteristic vector of the respiratory physiological index variable in the first respiratory characteristic matrix and the first reference respiratory characteristic vector.

The respiratory physiological index variable set can be a full physiological database of the respiratory physiological side or a subset of the full physiological database of the respiratory physiological side.

The breathing characteristic vector of the target index is corresponded to the first breathing characteristic matrix to obtain a first reference breathing characteristic vector, and specifically, convolution operation and pooling operation can be performed on the breathing characteristic vector of the target index to obtain the first reference breathing characteristic vector corresponding to the first breathing characteristic matrix.

The matching degree may be a euclidean distance, a cosine distance, a manhattan distance, and the like, which is not limited in this embodiment. The breathing physiological index variable with the matching degree smaller than the preset distance can be determined as the pre-stored breathing physiological index variable, the preset distance can be set according to actual conditions, and the embodiment does not limit the pre-stored breathing physiological index variable.

In the optimization process of the preset second respiratory state identification model, the second confidence coefficient of the pre-stored respiratory physiological index variable close to the application scene matching degree in the mapping data (namely the corresponding optimized sample respiratory physiological data) is increased.

Optionally, in some embodiments, the step of "calculating a second cost function between the predicted sample respiratory physiological data of the same optimized sample respiratory state data and the corresponding optimized sample respiratory physiological data" may include:

calculating a prestored respiration physiological index variable corresponding to a target index of the same optimized sample respiration state data, and calculating a first homogeneous coefficient between comparison indexes corresponding to the target index in the optimized sample respiration physiological data corresponding to the optimized sample respiration state data;

according to the first homogeneity coefficient and the second confidence coefficient, calculating index difference cost between a prestored respiration physiological index variable corresponding to each target index in the optimized sample respiration state data and a comparison index;

and executing weighted average operation on the index difference cost corresponding to each target index to obtain a second cost function between the predicted sample respiratory physiological data of the optimized sample respiratory state data and each corresponding optimized sample respiratory physiological data.

The target index corresponds to a plurality of pre-stored respiratory physiological index variables, and the first homogeneity coefficient comprises a homogeneity coefficient between each pre-stored respiratory physiological index variable corresponding to the target index and a corresponding comparison index.

The first homogeneity coefficient may be specifically measured by a vector distance, which specifically refers to a vector distance between a respiratory feature vector of the pre-stored respiratory physiological index variable and a respiratory feature vector of the comparison index. The larger the vector distance is, the smaller the first homogeneous coefficient is; the smaller the vector distance, the larger the first homogenous coefficient. The vector distance may be specifically an euclidean distance, a cosine distance, or the like.

The weighted average operation is performed on the index difference cost corresponding to each target index, specifically, the index difference cost corresponding to each target index is weighted and summed.

(2) Pre-training of the first respiratory state recognition model.

If the first respiratory state identification model is pre-trained by the conversion device according to artificial intelligence, before the step "convert the original sample respiratory physiological data through the first respiratory state identification model to obtain the converted predicted sample respiratory state data", the following example is further provided:

acquiring assisted respiration data pairs, wherein one assisted respiration data pair comprises target sample respiration physiological data and a plurality of corresponding optimized sample respiration state data, and the optimized sample respiration state data is reference conversion of the target sample respiration physiological data;

converting the target sample respiratory physiological data through a preset first respiratory state identification model to obtain converted prediction optimization sample respiratory state data;

calculating a third cost function between the respiratory state data of the prediction optimization sample of the respiratory physiological data of the same target sample and the respiratory state data of each corresponding optimization sample;

and optimizing a model framework of a preset first respiratory state identification model according to a third cost function to obtain the first respiratory state identification model.

The model architecture of the preset first respiratory state identification model can be optimized by adopting a back propagation algorithm or a random gradient descent algorithm, and the model architecture of the preset first respiratory state identification model is optimized according to a third cost function, so that the third cost function is smaller than the preset cost function, and the first respiratory state identification model is obtained. The preset cost function can be set according to actual conditions.

Optionally, in some embodiments, the step of "converting the target sample respiratory physiological data by presetting the first respiratory state identification model to obtain the converted prediction optimization sample respiratory state data" may include:

acquiring a respiration characteristic vector corresponding to each target index in the target sample respiration physiological data through a preset first respiration state identification model;

predicting at least two pre-stored respiration state index variables corresponding to each target index and a third confidence coefficient of each target index converted into the corresponding pre-stored respiration state index variable according to the respiration feature vector;

according to the third confidence coefficient, determining target respiratory state index variables corresponding to each target index from prestored respiratory state index variables;

and performing weighted average operation on each target respiration state index variable to obtain the prediction optimization sample respiration state data after the target sample respiration physiological data is converted.

For each target index, the prestored breathing state index variable with the maximum third confidence coefficient may be used as the target breathing state index variable corresponding to the target index, and the target breathing state index variable may also be regarded as the conversion result of the target index. The fusion mode of the target respiratory state index variables may specifically be to splice the target respiratory state index variables in a certain mode to obtain the predictive optimization sample respiratory state data.

Optionally, in some embodiments, the step of "predicting, according to the respiratory feature vector, at least two pre-stored respiratory state index variables corresponding to each target index, and converting each target index into a third confidence of the corresponding pre-stored respiratory state index variable" may include:

acquiring a respiratory state index variable set, wherein the respiratory state index variable set comprises a plurality of respiratory state index variables;

the method comprises the steps that a breathing characteristic vector of a target index corresponds to a second breathing characteristic matrix through the matching relation between a first breathing characteristic matrix on a breathing physiology side and a second breathing characteristic matrix on a breathing state side in a preset first breathing state recognition model, and a second reference breathing characteristic vector is obtained;

and determining at least two pre-stored respiratory state index variables corresponding to each target index in the respiratory state index variables and a third confidence coefficient of each target index converted into the corresponding pre-stored respiratory state index variable according to the matching degree of the respiratory feature vector of the respiratory state index variable in the second respiratory feature matrix and the second reference respiratory feature vector.

The set of respiratory state index variables may be a full physiological database of the respiratory state side, or a subset of the full physiological database of the respiratory state side.

The breathing characteristic vector of the target index is corresponded to the second breathing characteristic matrix to obtain a second reference breathing characteristic vector, and specifically, convolution operation and pooling operation can be performed on the breathing characteristic vector of the target index to obtain the second reference breathing characteristic vector corresponding to the second breathing characteristic matrix.

The matching degree may be a euclidean distance, a cosine distance, a manhattan distance, and the like, which is not limited in this embodiment. The breathing state index variable with the matching degree smaller than the preset distance can be determined as the pre-stored breathing state index variable, the preset distance can be set according to actual conditions, and the embodiment does not limit the pre-stored breathing state index variable.

In the optimization process of the preset first respiratory state identification model, a third confidence coefficient of a prestored respiratory state index variable close to the application scene matching degree in the mapping data (namely, the corresponding optimized sample respiratory state data) is increased.

Optionally, in some embodiments, the step of "calculating a third cost function between the predicted optimized sample respiratory state data of the same target sample respiratory physiological data and the corresponding optimized sample respiratory state data" may include:

calculating a prestored respiratory state index variable corresponding to a target index of the same target sample respiratory physiological data, and calculating a second homogeneity coefficient between comparison indexes corresponding to the target index in optimized sample respiratory state data corresponding to the target sample respiratory physiological data;

according to the second homogeneity coefficient and the third confidence coefficient, calculating index difference cost between a pre-stored respiratory state index variable corresponding to each target index in the target sample respiratory physiological data and a comparison index;

and executing weighted average operation on the index difference cost corresponding to each target index to obtain a third price function between the predicted optimized sample respiratory state data of the target sample respiratory physiological data and the corresponding optimized sample respiratory state data.

The target index corresponds to a plurality of pre-stored respiratory state index variables, and the second homogeneity coefficient comprises a homogeneity coefficient between each pre-stored respiratory state index variable corresponding to the target index and the corresponding comparison index.

The second homogeneity coefficient may specifically be measured by a vector distance, where the vector distance specifically refers to a vector distance between a respiratory feature vector of the pre-stored respiratory state index variable and a respiratory feature vector of the comparison index. The larger the vector distance is, the smaller the second homogeneity coefficient is; the smaller the vector distance, the larger the second homogeneity coefficient. The vector distance may be specifically an euclidean distance, a cosine distance, or the like.

The weighted average operation is performed on the index difference costs corresponding to each target index, which may be specifically to perform weighted summation on the index difference costs corresponding to each target index.

For a specific calculation method of the third cost function, reference may be made to the description of the related embodiment (calculation of the second cost function) in the training of the second respiratory state identification model, and details are not repeated here.

As can be seen from the above, the embodiment may acquire the physiological parameter of respiration to be evaluated, where the physiological parameter of respiration includes at least one physiological index of respiration; acquiring a respiratory characteristic vector corresponding to at least one respiratory physiological index in respiratory physiological parameters through a first respiratory state identification model, wherein the first respiratory state identification model is obtained by training according to a plurality of sample respiratory data pairs, each sample respiratory data pair comprises sample respiratory state data and sample respiratory physiological data obtained by converting the sample respiratory state data through a second respiratory state identification model, the second respiratory state identification model is obtained by training according to optimized respiratory data pairs, one optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data, and the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data; predicting at least two pre-stored respiratory state indexes corresponding to each respiratory physiological index and a first confidence coefficient of each respiratory physiological index converted into the corresponding pre-stored respiratory state index through a first respiratory state recognition model according to the respiratory feature vector; determining a predicted respiratory state index corresponding to each respiratory physiological index from prestored respiratory state indexes according to a first confidence coefficient through a first respiratory state identification model; and performing weighted average operation on each predicted respiratory state index through the first respiratory state identification model to obtain the respiratory state parameter after the respiratory physiological parameter is converted. The embodiment of the application can train the second respiratory state recognition model according to a plurality of optimized sample respiratory physiological data, increases the diversity of constructed sample respiratory data pairs, provides richer preset respiratory state data for the training of the first respiratory state recognition model, and further improves the conversion quality.

According to the method described in the foregoing embodiment, the following will be described in further detail by way of example in which the conversion apparatus according to artificial intelligence is specifically integrated in a server, which may specifically be a cloud server or the like.

In order to more clearly describe the scheme provided by the embodiment of the present application, the embodiment of the present application further provides the following examples:

201. the server receives the respiratory physiological parameters to be evaluated, which are sent by the terminal, wherein the respiratory physiological parameters comprise at least one respiratory physiological index.

202. The server obtains a respiratory characteristic vector corresponding to at least one respiratory physiological index in the respiratory physiological parameters through a first respiratory state recognition model, wherein the first respiratory state recognition model is obtained by training according to a plurality of sample respiratory data pairs, each sample respiratory data pair comprises sample respiratory state data and sample respiratory physiological data obtained by converting the sample respiratory state data through a second respiratory state recognition model, the second respiratory state recognition model is obtained by training according to optimized respiratory data pairs, each optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data, and the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data.

In the present embodiment, the first respiratory state recognition model and the second respiratory state recognition model may be NMT models according to a recurrent neural network, a convolutional neural network, and self-attention, NMT models using RNN, CNN, and self-attention in a mixed manner, or the like.

The second respiratory state identification model is obtained by training according to the optimized respiratory data pair, one optimized respiratory data pair comprises one optimized sample respiratory state data and a plurality of corresponding optimized sample respiratory physiological data, namely the second respiratory state identification model is obtained by training according to a plurality of mapping data. The trained second respiratory state recognition model converts the sample respiratory state data, and can output more diversified respiratory physiological side corresponding user states.

203. The server predicts at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes according to the respiratory characteristic vectors through the first respiratory state recognition model, and converts the respiratory physiological indexes into first confidence coefficients of the corresponding pre-stored respiratory state indexes.

204. And the server determines a predicted respiratory state index corresponding to each respiratory physiological index from the prestored respiratory state indexes according to the first confidence coefficient through the first respiratory state identification model.

205. And the server executes weighted average operation on each predicted respiratory state index through the first respiratory state identification model to obtain the respiratory state parameters after the respiratory physiological parameters are converted.

206. And the server sends the breathing state parameters to the terminal.

The first respiratory state recognition model and the second respiratory state recognition model are trained, and the second respiratory state recognition model is obtained by training according to a plurality of mapping data, so that sample respiratory physiological data in a sample respiratory data pair constructed by the second respiratory state recognition model are relatively more diverse; the first respiratory state recognition model is obtained by training sample respiratory data constructed according to the second respiratory state recognition model, and the diversity of the respiratory state parameters of a preset user in the sample respiratory data pair can be increased for the first respiratory state recognition model, so that the conversion quality of the first respiratory state recognition model is enhanced through richer respiratory state parameters.

After receiving the breathing state parameters, the terminal can display the converted breathing state parameters on a display of the electronic equipment.

In a specific embodiment, the first respiratory state recognition model is used as a forward respiratory state recognition model, the second respiratory state recognition model is used as a reverse respiratory state recognition model, the forward respiratory state recognition model converts source data content into target data content, the reverse respiratory state recognition model converts the target data content into the source data content, and the specific training processes of the forward respiratory state recognition model and the reverse respiratory state recognition model are as follows:

2001. acquiring an initial respiratory data pair, which specifically comprises source data content K0 and target data content H0;

2002. selecting a training direction, wherein a scheme for increasing the diversity of identification data of the forward breathing state identification model can be selected, and a scheme for improving the conversion quality of the forward breathing state identification model by improving a reverse conversion technology can also be selected;

2003. if the scheme for increasing the diversity of the identification data of the forward breathing state identification model is selected in step 2002, the forward breathing state identification model may be trained according to a training target driven by diversity, specifically, various target end candidate identification data may be obtained, both the target end candidate identification data and the target data content H0 may be used as target data mapping data of the source data content K0 (which may be considered as optimized sample breathing state data mentioned in the foregoing embodiment), and the forward breathing state identification model may be trained according to the target data mapping data and the source data content K0;

2004. according to the training in the step 2003, optimizing a model framework of the forward respiration state recognition model so as to enable a cost function between the converted prediction optimization sample respiration state data and the target data mapping data to meet a preset condition, and obtaining a trained forward respiration state recognition model M, wherein the forward respiration state recognition model M can increase the diversity of recognition data;

2005. if the scheme for improving the conversion quality of the forward breathing state recognition model by improving the reverse conversion technology is selected in step 2002, the reverse breathing state recognition model can be trained according to a training target driven by diversity, specifically, various source candidate recognition data can be obtained, both the source candidate recognition data and the source data content K0 can be used as source data mapping data (which can be considered as optimized sample breathing physiological data mentioned in the previous embodiment) of the target data content H0, and the reverse breathing state recognition model is trained according to the source data mapping data and the target data content H0;

2006. according to the training in the step 2005, optimizing a model architecture of a reverse respiration state recognition model so that a cost function between the converted prediction sample respiration physiological data and source data mapping data meets a preset condition, and obtaining a trained reverse respiration state recognition model;

2007. acquiring single target respiratory state data H1;

2008. according to the reverse breathing state recognition model trained in the step 2006, single target breathing state data H1 is converted to obtain pseudo source data content K1, the pseudo source data content K1 and the single target breathing state data H1 form a sample breathing data pair, and the reverse breathing state recognition model is trained according to a plurality of source data mapping data, so that the diversity of the constructed pseudo source data content K1 can be increased, and richer preset breathing state data are provided;

2009. combining the sample respiration data pair with the initial respiration data pair to obtain a sample training set of the forward respiration state recognition model, wherein the sample training set comprises source data content (K0 + K1) and target data content (H0 + H1);

2010. the method comprises the steps that source data content (K0 + K1) can be used as input of a forward respiration state recognition model to obtain converted predicted sample respiration state data, and conversion training from source data to target data is conducted on the forward respiration state recognition model according to cross entropy of maximum likelihood estimation between the predicted sample respiration state data and corresponding target data content (H0 + H1);

2011. optimizing a model architecture of the forward respiration state recognition model through the training in the step 2010 so that the cross entropy of the maximum likelihood estimation meets a preset condition to obtain a trained forward respiration state recognition model M1; because the reverse respiration state recognition model has stronger capability of outputting various recognition data, the source end of the combined sample training set contains richer information, and the conversion quality of the final forward respiration state recognition model M1 can be improved.

Optionally, in some embodiments, training for increasing the diversity of the identification data may be performed only on the forward respiration state identification model (see steps 2001-2004), or only by improving the reverse conversion technique, the conversion quality of the forward respiration state identification model may be enhanced (see steps 2001, 2002, and steps 2005-2011).

Optionally, in other embodiments, the forward respiration state recognition model M1 may be obtained by training on the basis of the forward respiration state recognition model M obtained in step 2004, that is, the forward respiration state recognition model is trained twice; specifically, training of increasing identification data diversity can be performed on the forward respiration state identification model and the reverse respiration state identification model to obtain a trained forward respiration state identification model M and a trained reverse respiration state identification model, then a sample respiration data pair is obtained through the trained reverse respiration state identification model, and conversion training from source data to target data is performed on the forward respiration state identification model M to obtain a forward respiration state identification model M1.

In current related art, training of respiratory state recognition models typically optimizes the model architecture for the library using initial respiratory data with a single reference, which limits the use of resources and causes the generated transitions to blindly and unreasonably approximate the reference transitions.

The embodiment of the application can train the reverse respiration state recognition model according to diversity-driven training targets to increase the diversity of the preset user respiration state parameters of the constructed training set, thereby enhancing the forward respiration state recognition model by enriching the source data information in the training set, combining with the cross entropy loss of the neural network machine respiration state recognition model NMT and finally improving the conversion quality of the forward respiration state recognition model.

The application can be used for any neural network machine conversion system (such as RNNsearch, transformer, etc.) realized according to any deep learning framework.

As can be seen from the above, in this embodiment, the server may receive the physiological parameter of respiration to be evaluated, which is sent by the terminal, where the physiological parameter of respiration includes at least one physiological index of respiration; acquiring a respiratory characteristic vector corresponding to at least one respiratory physiological index in the respiratory physiological parameters through a first respiratory state recognition model, wherein the first respiratory state recognition model is obtained by training according to a plurality of sample respiratory data pairs, each sample respiratory data pair comprises sample respiratory state data and sample respiratory physiological data obtained by converting the sample respiratory state data through a second respiratory state recognition model, the second respiratory state recognition model is obtained by training according to optimized respiratory data pairs, one optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data, and the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data; the server predicts at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes and a first confidence coefficient of converting the respiratory physiological indexes into the corresponding pre-stored respiratory state indexes according to the respiratory feature vector through the first respiratory state recognition model; determining a predicted respiratory state index corresponding to each respiratory physiological index from prestored respiratory state indexes according to a first confidence coefficient through a first respiratory state identification model; executing weighted average operation on each predicted respiratory state index through a first respiratory state identification model to obtain respiratory state parameters after respiratory physiological parameter conversion; and the server sends the breathing state parameters to the terminal. The embodiment of the application can train the second respiratory state recognition model according to a plurality of optimized sample respiratory physiological data, increases the diversity of constructed sample respiratory data pairs, provides richer preset respiratory state data for the training of the first respiratory state recognition model, and further improves the conversion quality.

Referring to fig. 2, fig. 2 is a schematic block diagram of a monitoring system 110 of a respiratory function rehabilitation training device based on artificial intelligence according to an embodiment of the present invention, including:

a response module 1101, configured to establish a communication connection with the respiratory function rehabilitation training instrument in response to the respiratory function rehabilitation monitoring request.

The acquiring module 1102 is configured to acquire a user physiological parameter from the respiratory function rehabilitation training instrument, where the user physiological parameter is acquired by a sensor configured in the respiratory function rehabilitation training instrument.

The monitoring module 1103 is configured to obtain a respiratory physiological parameter to be evaluated from the user physiological parameter, and call the first respiratory state identification model to perform processing to obtain a respiratory state parameter after the respiratory physiological parameter is converted; and evaluating according to the respiratory state parameters by combining a preset respiratory function rehabilitation training table to obtain a monitoring result of the respiratory function rehabilitation training instrument.

It should be noted that, as for the implementation principle of the monitoring system 110 for the respiratory function rehabilitation training apparatus based on artificial intelligence, reference may be made to the implementation principle of the monitoring method for the respiratory function rehabilitation training apparatus based on artificial intelligence, which is not described herein again. It should be understood that the division of the modules of the above apparatus is only a logical division, and the actual implementation may be wholly or partially integrated into one physical entity, or may be physically separated. And these modules can be realized in the form of software called by processing element; or may be implemented entirely in hardware; and part of the modules can be realized in the form of calling software by the processing element, and part of the modules can be realized in the form of hardware.

For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or one or more microprocessors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), among others. For another example, when some of the above modules are implemented in the form of a processing element scheduler code, the processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. As another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

The embodiment of the present invention provides a computer device 100, where the computer device 100 includes a processor and a non-volatile memory storing computer instructions, and when the computer instructions are executed by the processor, the computer device 100 executes the monitoring system 110 of the artificial intelligence based respiratory function rehabilitation training instrument. As shown in fig. 3, fig. 3 is a block diagram of a computer device 100 according to an embodiment of the present invention. The computer device 100 comprises a monitoring system 110 of an artificial intelligence based breathing function rehabilitation training instrument, a memory 111, a processor 112 and a communication unit 113.

To facilitate the transfer or interaction of data, the elements of the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other, directly or indirectly. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The monitoring system 110 of the artificial intelligence based respiratory function rehabilitation training device includes at least one software function module which can be stored in a memory 111 in the form of software or firmware (firmware) or solidified in an Operating System (OS) of the computer device 100. The processor 112 is used for executing the monitoring system 110 of the artificial intelligence based respiratory function rehabilitation training device stored in the memory 111, such as a software function module and a computer program included in the monitoring system 110 of the artificial intelligence based respiratory function rehabilitation training device.

The embodiment of the invention provides a readable storage medium, which comprises a computer program, and when the computer program runs, the computer device where the readable storage medium is located is controlled to execute the monitoring method of the artificial intelligence based respiratory function rehabilitation training instrument.

The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical applications, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated. The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical applications, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated.

Claims

1. The monitoring system of the respiratory function rehabilitation training instrument based on artificial intelligence is characterized by comprising a server and the respiratory function rehabilitation training instrument in communication connection with the server;

the server is used for responding to the respiratory function rehabilitation monitoring request and establishing communication connection with the respiratory function rehabilitation training instrument; acquiring the physiological parameters of the user from the respiratory function rehabilitation training instrument; acquiring a respiratory physiological parameter to be evaluated from the user physiological parameter, and calling a first respiratory state identification model for processing to obtain a respiratory state parameter after the respiratory physiological parameter is converted; evaluating according to the respiratory state parameters by combining a preset respiratory function rehabilitation training table to obtain a monitoring result of the respiratory function rehabilitation training instrument; determining a plurality of target users with continuous abnormal breathing state characteristics based on the monitoring result of the breathing function training instrument; determining target on-line pushing information matched with each target user according to the continuous abnormal breathing state characteristics of each target user, wherein the target on-line pushing information comprises pushing information of medical equipment products and pushing information of rehabilitation training information.

2. The system of claim 1, wherein the server is further configured to:

acquiring the respiratory physiological parameter to be evaluated, wherein the respiratory physiological parameter comprises at least one respiratory physiological index;

acquiring a respiratory feature vector corresponding to the at least one respiratory physiological index in the respiratory physiological parameter through a first respiratory state identification model, wherein the first respiratory state identification model is obtained by training according to a plurality of sample respiratory data pairs, each sample respiratory data pair comprises sample respiratory state data and sample respiratory physiological data obtained by converting the sample respiratory state data through a second respiratory state identification model, the second respiratory state identification model is obtained by optimizing a model architecture of a preset second respiratory state identification model according to optimized respiratory data pairs, one optimized respiratory data pair comprises optimized sample respiratory state data and a plurality of optimized sample respiratory physiological data, the optimized sample respiratory physiological data are mapping data of the optimized sample respiratory state data, and the model architecture of the preset second respiratory state identification model is optimized according to a conversion result of the preset second respiratory state identification model on the optimized sample respiratory state data and a second generation cost function between the optimized sample respiratory physiological data corresponding to the optimized sample respiratory state data;

predicting at least two pre-stored respiratory state indexes corresponding to each respiratory physiological index and a first confidence coefficient of each respiratory physiological index converted into a corresponding pre-stored respiratory state index through the first respiratory state recognition model according to the respiratory feature vector;

determining a predicted respiratory state index corresponding to each respiratory physiological index from the prestored respiratory state indexes according to the first confidence coefficient through the first respiratory state identification model;

and performing weighted average operation on each predicted respiratory state index through the first respiratory state identification model to obtain the respiratory state parameter after the respiratory physiological parameter is converted.

3. The system of claim 2, wherein the server is further configured to:

corresponding the breathing characteristic vector to a second breathing characteristic matrix through a matching relation between a first breathing characteristic matrix at a breathing physiology side and a second breathing characteristic matrix at a breathing state side in the first breathing state identification model to obtain a reference breathing characteristic vector;

and determining at least two pre-stored respiratory state indexes corresponding to the respiratory physiological indexes in the respiratory physiological index set according to the matching degree of the respiratory physiological indexes of the respiratory state side in the second respiratory characteristic matrix and the reference respiratory characteristic vector, and converting the respiratory physiological indexes into first confidence degrees of the corresponding pre-stored respiratory state indexes.

4. The system of claim 2, wherein the server is further configured to:

acquiring an optimized respiration data pair;

corresponding the breathing characteristic vector of the target index to the first breathing characteristic matrix through the matching relation between the second breathing characteristic matrix at the breathing state side and the first breathing characteristic matrix at the breathing physiology side in the preset second breathing state identification model to obtain a first reference breathing characteristic vector;

determining at least two pre-stored respiratory physiological index variables corresponding to each target index in the respiratory physiological index variables and a second confidence coefficient of each target index converted into a corresponding pre-stored respiratory physiological index variable according to the respiratory characteristic vector of the respiratory physiological index variable in the first respiratory characteristic matrix and the matching degree of the first reference respiratory characteristic vector;

according to the second confidence coefficient, determining a target respiration physiological index variable corresponding to each target index from the prestored respiration physiological index variables;

performing weighted average operation on each target respiration physiological index variable to obtain predicted sample respiration physiological data after the optimized sample respiration state data is converted;

according to the first similarity coefficient and the second confidence coefficient, calculating index difference cost between prestored respiratory physiological index variables corresponding to the target indexes in the optimized sample respiratory state data and comparison indexes;

performing weighted average operation on the index difference cost corresponding to each target index to obtain a second cost function between the predicted sample respiratory physiological data of the optimized sample respiratory state data and the corresponding optimized sample respiratory physiological data;

optimizing a model framework of a preset second respiratory state recognition model according to the second cost function to obtain a second respiratory state recognition model;

converting the sample respiratory state data through a second respiratory state identification model to obtain converted sample respiratory physiological data, and combining the converted sample respiratory physiological data with the corresponding sample respiratory state data to form a sample respiratory data pair;

determining the sample respiration data pair and the initial respiration data pair as a sample training set of the first respiration state identification model, wherein the initial sample respiration physiological data and the converted sample respiration physiological data are original sample respiration physiological data, and the initial sample respiration state data and the sample respiration state data are original sample respiration state data;

and converting the original sample respiratory physiological data through a first respiratory state identification model to obtain converted predicted sample respiratory state data, and performing conversion training from a respiratory physiological side to a respiratory state side on the first respiratory state identification model according to a first cost function between the predicted sample respiratory state data and the original sample respiratory state data corresponding to the original sample respiratory physiological data.

5. The system of claim 4, wherein the server is further configured to:

acquiring assisted respiration data pairs, wherein one assisted respiration data pair comprises a target sample respiration physiological data and a plurality of corresponding optimized sample respiration state data, and the optimized sample respiration state data is a reference conversion of the target sample respiration physiological data;

calculating a third cost function between the predicted optimized sample respiratory state data of the same target sample respiratory physiological data and the corresponding optimized sample respiratory state data;

and optimizing a model framework of a preset first respiratory state identification model according to the third cost function to obtain the first respiratory state identification model.

6. The system of claim 5, wherein the server is further configured to:

acquiring a respiratory characteristic vector corresponding to each target index in the respiratory physiological data of the target sample through a preset first respiratory state identification model;

predicting at least two pre-stored respiratory state index variables corresponding to each target index according to the respiratory feature vector corresponding to the target index, and converting each target index into a third confidence coefficient of the corresponding pre-stored respiratory state index variable;

according to the third confidence coefficient, determining a target respiratory state index variable corresponding to each target index from the prestored respiratory state index variables;

and performing weighted average operation on each target respiratory state index variable to obtain the prediction optimization sample respiratory state data after the target sample respiratory physiological data is converted.

7. The system of claim 6, wherein the server is further configured to:

corresponding the breathing characteristic vector of the target index to a second breathing characteristic matrix through a matching relation between a first breathing characteristic matrix at a breathing physiology side and a second breathing characteristic matrix at a breathing state side in the preset first breathing state identification model to obtain a second reference breathing characteristic vector;

and determining at least two pre-stored respiratory state index variables corresponding to each target index in the respiratory state index variables and a third confidence coefficient of each target index converted into a corresponding pre-stored respiratory state index variable according to the respiratory feature vector of the respiratory state index variable in the second respiratory feature matrix and the matching degree of the second reference respiratory feature vector.

8. The system of claim 7, wherein the server is further configured to:

calculating a prestored breathing state index variable corresponding to a target index of the same target sample breathing physiological data, and calculating a second homogeneity coefficient between comparison indexes corresponding to the target index in optimized sample breathing state data corresponding to the target sample breathing physiological data;

according to the second homogeneity coefficient and the third confidence coefficient, calculating index difference cost between a pre-stored respiration state index variable corresponding to each target index in the target sample respiration physiological data and a comparison index;

and performing weighted average operation on the index difference cost corresponding to each target index to obtain a third price function between the predicted optimized sample respiratory state data of the target sample respiratory physiological data and the corresponding optimized sample respiratory state data.

9. The system of claim 4, wherein the server is further configured to:

predicting at least two undetermined sample respiratory physiological indexes corresponding to the sample respiratory physiological indexes according to the respiratory feature vectors, and converting each sample respiratory physiological index into a fourth confidence coefficient of the corresponding undetermined sample respiratory physiological index;

according to the fourth confidence coefficient, determining target indexes corresponding to the respiratory physiological indexes of the samples from the respiratory physiological indexes of the samples to be determined, and performing weighted average operation on the target indexes to obtain predicted sample respiratory state data;

and optimizing the model framework of the first respiratory state recognition model according to the first cost function to obtain the trained first respiratory state recognition model.

10. A monitoring method of a respiratory function rehabilitation training instrument based on artificial intelligence is characterized by comprising the following steps:

acquiring a user physiological parameter from the respiratory function rehabilitation training instrument, wherein the user physiological parameter is acquired by a sensor configured for the respiratory function rehabilitation training instrument;

acquiring a respiratory physiological parameter to be evaluated from the user physiological parameter, and calling a first respiratory state identification model for processing to obtain a respiratory state parameter after the respiratory physiological parameter is converted;

obtaining a monitoring result of the respiratory function training instrument;

determining target on-line pushing information matched with each target user according to the continuous abnormal breathing state characteristics of each target user, wherein the target on-line pushing information comprises pushing information of medical equipment products and pushing information of rehabilitation training information.