Model training method and device, physiological index detection method and device and electronic equipment

Abstract

The embodiment of the invention provides a model training method, a physiological index detection method and device, electronic equipment and a computer readable storage medium, and relates to the field of artificial intelligence. The method comprises the following steps: training a first initial model by using a plurality of first training samples to obtain a first index prediction model; training a second initial model corresponding to the partial types of PPG signals in a knowledge distillation mode by combining a plurality of second training samples corresponding to the same partial types of PPG signals with the first index prediction model to obtain a corresponding second index prediction model, and training a third initial model by a plurality of third training samples to obtain a third index prediction model; and obtaining a regression model. According to the embodiment of the invention, the number of all kinds of PPG signals does not need to be limited, and the more the number is, the higher the accuracy is.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology. Specifically, the present application relates to a model training method, a physiological indicator detection method, a device, an electronic device, a computer-readable storage medium and a computer program product. Background Art

[0002] Smart wearable devices are widely used to monitor long-term or short-term health conditions.

[0003] Currently, many smart wearable devices use locally deployed neural network models to predict physiological indicators based on photoplethysmography (PPG) signals. Since there are many types of PPG signals, and different models of smart wearable devices from the same manufacturer have different hardware conditions for collecting PPG signals, different models of smart wearable devices can collect different numbers and types of PPG signals.

[0004] The related technology requires that for each model of smart wearable device, the PPG signals that can be collected by the model of smart wearable device be collected as training samples to train the corresponding neural network model, and the training efficiency is low. Summary of the invention

[0005] The embodiments of the present application provide a model training method, device, electronic device, computer-readable storage medium and computer program product, which can solve the above-mentioned problems of the prior art. The technical solution is as follows: According to a first aspect of an embodiment of the present application, a model training method is provided, which is applied to a server, and the method includes: Training a first initial model with a plurality of first training samples to obtain a first indicator prediction model, wherein the first training samples include all types of PPG signals within a time period, and the first training samples also include sample labels, and the sample labels are used to indicate a physiological indicator interval in which an actual physiological indicator value of the corresponding time period is located; Splitting each first training sample into a plurality of second training samples, each second training sample including a portion of the PPG signals in the corresponding first training sample, and the second training sample also including a training label that is the same as the training label of the corresponding first training sample; Using a plurality of second training samples corresponding to the same partial types of PPG signals, combined with the first indicator prediction model, a second initial model corresponding to the partial types of PPG signals is trained through knowledge distillation to obtain a corresponding second indicator prediction model, each second indicator prediction model is used to be deployed in a smart wearable device that collects a corresponding type of PPG signal; The third initial model is trained with multiple third training samples to obtain a regression model, wherein the third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​in the same time period.

[0006] According to a second aspect of an embodiment of the present application, a physiological index detection method is provided, which is applied to a smart wearable device, and the method includes: Collecting some types of PPG signals in the first period; If it is determined that the target indicator prediction model has not been calibrated for a preset period of time, all types of PPG signals in the second period are obtained, and all types of PPG signals in the second period are uploaded to the server, and the server is trained to obtain a first indicator prediction model, a regression model and a plurality of second indicator prediction models by the model training method according to claims 1-7, and the target indicator prediction model is one of the plurality of second indicator prediction models; receiving and displaying the predicted value of the first physiological indicator in the first time period returned by the server; The first physiological indicator prediction value is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

[0007] According to a third aspect of an embodiment of the present application, a model training device is provided, which is applied to a server, and the device includes: A first training module, configured to train a first initial model with a plurality of first training samples to obtain a first indicator prediction model, wherein the first training samples include all types of PPG signals within a time period, and the first training samples also include sample labels, wherein the sample labels are used to indicate a physiological indicator interval in which an actual physiological indicator value of a corresponding time period is located; A sample splitting module, used to split each first training sample into multiple second training samples, each second training sample includes a part of the types of PPG signals in the corresponding first training sample, and the second training sample also includes a training label and is the same as the training label of the corresponding first training sample; A second training module is used to train a second initial model corresponding to the partial types of PPG signals by combining a plurality of second training samples corresponding to the same partial types of PPG signals with the first indicator prediction model through knowledge distillation to obtain a corresponding second indicator prediction model; each second indicator prediction model is used to be deployed on a smart wearable device capable of collecting a corresponding type of PPG signal; The third training module is used to train the third initial model with multiple third training samples to obtain a regression model. The third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​in the same time period.

[0008] According to a fourth aspect of an embodiment of the present application, a physiological index detection device is provided, which is applied to a smart wearable device, and the device includes: A signal acquisition module, used for acquiring some types of PPG signals in a first period of time; A signal acquisition module, configured to acquire all types of PPG signals in a second period if it is determined that the target indicator prediction model has not been calibrated for a preset period of time, and upload all types of PPG signals in the second period to a server, wherein the server obtains a first indicator prediction model, a regression model, and a plurality of second indicator prediction models through training by the model training method provided in the first aspect, and the target indicator prediction model is one of the plurality of second indicator prediction models; An indicator display module, used for receiving and displaying the predicted value of the first physiological indicator of the first time period returned by the server; The first physiological indicator prediction value is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

[0009] According to a fifth aspect of an embodiment of the present application, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the method provided in the first aspect or the second aspect above.

[0010] According to a sixth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method provided in the first aspect or the second aspect are implemented.

[0011] According to a seventh aspect of an embodiment of the present application, a computer program product is provided, including a computer program, which implements the steps of the method provided in the first aspect or the second aspect when the computer program is executed by a processor.

[0012] The beneficial effects of the technical solution provided by the embodiment of the present application are: First, a first indicator prediction model is obtained by training based on a first training sample including all kinds of PPG signals, and then the first training sample is split to obtain multiple second training samples, each of which includes some kinds of PPG signals. In combination with the first indicator prediction model, a second initial model corresponding to the some kinds of PPG signals is trained through knowledge distillation. The second indicator prediction model obtained by training based on the second training sample can obtain the knowledge of the first indicator prediction model, and then a model with higher accuracy is predicted based on fewer kinds of PPG signals. In addition, this solution does not need to limit the number of all kinds of PPG signals, but will produce an effect of higher accuracy with more numbers. In addition, the trained second indicator prediction model can also theoretically cover all models of smart wearable devices on the market, and corresponding deployable second indicator prediction models can be found for various smart wearable devices that collect some kinds of PPG signals. In addition, since multiple third training samples are further generated by each second indicator prediction model, and the third initial model is trained based on the multiple third training samples to obtain a regression model, a foundation is laid for the subsequent calibration of the results of the second indicator prediction model deployed on the smart wearable device. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in describing the embodiments of the present application.

[0014] Figure 1 A schematic diagram of a system architecture for implementing a physiological index detection method provided in an embodiment of the present application; Figure 2 A flowchart of a model training method provided in an embodiment of the present application; Figure 3 A schematic diagram of splitting a first training sample into a plurality of second training samples to train the corresponding second initial models provided in an embodiment of the present application; Figure 4 A one-round iterative processing process of a second initial model provided in an embodiment of the present application; Figure 5 A schematic diagram of a training process of a second initial model provided in an embodiment of the present application; Figure 6 A flowchart of a model training method provided in an embodiment of the present application; Figure 7 A schematic diagram of a physiological index detection method provided in an embodiment of the present application; Figure 8 A schematic diagram of the structure of a model training device provided in an embodiment of the present application; Fig. 9 A schematic diagram of the structure of a physiological index detection device provided in an embodiment of the present application; Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0016] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "an" and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when we call an element "connected" or "coupled" to another element, the one element can be directly connected or coupled to the other element, or it can refer to that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used here may include wireless connection or wireless coupling. The term "and / or" used here indicates at least one of the items defined by the term, for example, "A and / or B" can be implemented as "A", or as "B", or as "A and B".

[0017] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0018] The model training method, physiological indicator detection method, device, electronic device, computer-readable storage medium and computer program product provided in this application are intended to solve the above technical problems of the prior art.

[0019] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present application and the technical effects produced by the technical solutions of the present application. It should be noted that the following embodiments can refer to, draw on or combine with each other, and the same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0020] Figure 1 The system architecture diagram for implementing the physiological index detection method provided in the embodiment of the present application is as shown in the figure. The system includes a server 101 and a smart wearable device 102.

[0021] The server 101 deploys a first initial model, a third initial model, and multiple second initial models. The training sample used by the server 101 to train the first initial model is the first training sample, which includes all types of PPG signals within a period of time, and the sample label is used to indicate the physiological index interval where the actual physiological index value of the corresponding period of time is located. The training sample used by the server 101 to train the second initial model is the second training sample, which includes some types of PPG signals within a period of time, and the training label is the same as the training label of the corresponding first training sample. It should be understood that the partial types of PPG signals in the embodiment of the present application are relative to the full types of PPG signals. If the number of types of full types of PPG signals is N, then the number of types of partial types of PPG signals is 1 to N-1. For example, if the full types of PPG signals include the following 7 types of PPG signals: Photoplethysmography C Infrared (PPGCIR), where “C” is the code for fingertips; Photoplethysmography C Red (PPGCR) of fingertips, where “C” is the code for fingertips; Photoplethysmography C Yellow (PPGCY), where “C” is the code for the fingertip; Photoplethysmography Green Band (PPGGB); Wrist photoplethysmography D Infrared (PPGDIR), where “D” is the code for wrist; Wrist photoplethysmography D Red (PPGDR), where “D” is the code for wrist, and Wrist photoplethysmography D Green (PPGDG), where "D" is the code for wrist.

[0022] Correspondingly, some types of PPG signals are at least one of the above 7 types of PPG signals.

[0023] In the embodiment of the present application, each second initial model is trained using a fixed second training sample of some types of PPG signals. Taking the example that each second training sample only includes one type of PPG signal, then based on the above-mentioned 7 types of PPG signals, the embodiment of the present application can train 7 second initial models, and each second initial model is trained based on the second training sample including one PPG signal.

[0024] The embodiment of the present application adopts the knowledge distillation (KD) method when training the second initial model. The core idea of ​​this method is to improve the performance of the student model without changing its structure by guiding the lightweight student model to "imitate" the teacher model (or ensemble of multiple models) with better performance and more complex structure. In the embodiment of the present application, the model structure of the first indicator prediction model is more complex, and the training samples required for training contain more information, so it is used as the teacher model when performing knowledge distillation, and the second initial model is the student model accordingly.

[0025] After the server obtains the second indicator prediction model according to the second initial model training, it further obtains multiple third training samples according to each second prepared prediction model. The third training samples include the middle values ​​of the physiological indicator intervals output by the multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period. The training labels of the third training samples are the actual physiological indicator values ​​of the same time period. The third initial model is trained with the multiple third training samples to obtain a regression model. The regression model can input the middle values ​​of the physiological indicator intervals predicted by the multiple second indicator prediction models to predict more accurate physiological indicator values.

[0026] It should be understood that, since the first training sample, the second training sample and the third training sample all include sample labels, when the training samples are used for training, it belongs to supervised learning.

[0027] The smart wearable device in the embodiment of the present application may be a smart watch, a smart bracelet, a smart ring, a smart blood pressure monitor, etc., and the embodiment of the present application does not make any specific limitation.

[0028] The present application embodiment provides a model training method, which is applied to a server, such as Figure 2 As shown, the method includes: S101. Train a first initial model using a plurality of first training samples to obtain a first indicator prediction model.

[0029] The first training sample of the embodiment of the present application includes all types of PPG signals in a time period, and the first training sample also includes a sample label, and the sample label is used to indicate the physiological index interval in which the actual physiological index value of the corresponding time period is located. Since the first training sample of the embodiment of the present application has a sample label, the first initial model is trained in a supervised learning manner.

[0030] The embodiment of the present application does not specifically limit the length of a time period, for example, it can be 15 seconds, 30 seconds, 60 seconds, etc.

[0031] The sample labels of the embodiments of the present application are not specific physiological indicator values, but the physiological indicator intervals where the physiological indicator values ​​are located, that is, the first initial model of the present application is a classification model for performing classification tasks. The embodiments of the present application can characterize different physiological indicator intervals with different classification results. Taking systolic blood pressure as an example, the general range is 70~159mmhg. The embodiments of the present application can use a difference of 9mmhg as an interval, and set 10 intervals, where interval 1 is 70~79mmhg, interval 2 is 80~89mmhg, and so on. The 10 intervals correspond to 10 label categories. For example, they can be simply set to label 1~label 10, where label 1 represents the systolic blood pressure range of interval 1, and interval 10 represents the systolic blood pressure range of interval 10.

[0032] It can be understood that the obtained first indicator prediction model has the ability to input all types of PPG signals of a time period and output the category of the physiological indicator interval in which the physiological indicator value of the time period is located.

[0033] The type of the physiological indicator in the embodiment of the present application is at least one of diastolic blood pressure, systolic blood pressure, blood oxygen, and blood sugar.

[0034] S102: Split each first training sample into multiple second training samples.

[0035] The embodiment of the present application further splits each first training sample into multiple second training samples, each second training sample includes some types of PPG signals in the corresponding first training sample, and the second training sample also includes a training label and is the same as the training label of the corresponding first training sample. It should be understood that for multiple second training samples split from a first training sample, the PPG signals included in any two second training samples have at least one type of PPG signal difference. For example, a first training sample includes three types of PPG signals - PPGCR, PPGCB and PPGDR. If each split second training sample includes two PPG signals, then a second training sample including PPGCR and PPGCB, a second training sample including PPGCB and PPGDR, and a second training sample including PPGCR and PPGDR can be split.

[0036] It should be noted that, in the embodiment of the present application, the number of second training samples split out from each first training sample is consistent with the number of second initial models. For example, taking the example that the first training sample includes 7 types of PPG signals and each second training sample includes 1 type of PPG signal, then one first training sample can be split into 7 second training samples, and there are correspondingly 7 second initial models.

[0037] S103. Using a plurality of second training samples corresponding to the same partial type of PPG signal and combining with the first indicator prediction model, a second initial model corresponding to the partial type of PPG signal is trained through knowledge distillation to obtain a corresponding second indicator prediction model.

[0038] It should be noted that, in the embodiments of the present application, each second initial model needs to be trained based on a second training sample including a fixed PPG signal of any type. For example, a second initial model 1 needs to be trained based on a second training sample including only two types of PPG signals, PPGCR and PPGCY.

[0039] See also Figure 3 , which exemplarily shows a schematic diagram of an embodiment of the present application splitting a first training sample into multiple second training samples to train the corresponding second initial model. The figure exemplarily shows two first training samples, namely first training sample 1 and first training sample 2, each of which includes three types of PPG signals. The present application splits each first training sample into three second training samples. Therefore, the second initial model 1 is finally trained based on the second training samples 1 and 4, the second initial model 2 is trained based on the second training samples 2 and 5, and the second initial model 3 is trained based on the second training samples 3 and 6.

[0040] It should be understood that, since the label of the second training sample is the same as the training label of the corresponding first training sample, the training of the second initial model is also a supervised learning method and also belongs to a classification model.

[0041] Since the types of PPG signals input to the second indicator prediction model of the embodiment of the present application are fewer than those input to the first indicator prediction model, the embodiment of the present application can deploy the second indicator prediction model in a smart wearable device that collects PPG signals of corresponding types and ensure a high accuracy rate.

[0042] The first indicator prediction model of the embodiment of the present application only imparts knowledge to the second initial model and can be used in the subsequent calibration process. That is, in the calibration process, only the second indicator prediction models and regression models are used, and the first indicator prediction model is not needed. However, in other embodiments, the more powerful prediction ability of the first indicator prediction model can also be used to calibrate in combination with the second indicator prediction models and regression models.

[0043] In some embodiments, the embodiments of the present application can train multiple second indicator prediction models for each partial type of signal, and deploy the second indicator prediction model with the best performance in the smart wearable device. For example, five second initial models with different model structures are trained with multiple second training samples including PPGCR signals to obtain five second indicator prediction models. The model performance of the five second indicator prediction models is verified by setting a verification set, and the second indicator prediction model with the best model performance is deployed in the smart wearable device that can collect PPGCR signals.

[0044] S104: Train a third initial model using a plurality of third training samples to obtain a regression model.

[0045] After training each second indicator prediction model, the embodiment of the present application obtains all types of PPG signals within the same time period (the time period may be different from the time period in the first training sample), and then splits them into third training samples corresponding to each second indicator prediction model, respectively. The third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​of the same time period, so that the trained regression model can input the middle values ​​of multiple physiological indicator intervals, and output the predicted physiological indicator values ​​through integration processing.

[0046] The model training method provided in the embodiment of the present application first obtains a first indicator prediction model based on a first training sample including all kinds of PPG signals, and then splits the first training sample to obtain multiple second training samples, each of which includes some kinds of PPG signals. In combination with the first indicator prediction model, the second initial model corresponding to the some kinds of PPG signals is trained by knowledge distillation. The second indicator prediction model obtained by training based on the second training sample can obtain the knowledge of the first indicator prediction model, and then a model with higher accuracy is predicted based on fewer kinds of PPG signals. In addition, this solution does not need to limit the number of all kinds of PPG signals, but will produce an effect of higher accuracy with more numbers. In addition, the trained second indicator prediction model can also theoretically cover all models of smart wearable devices on the market, and corresponding deployable second indicator prediction models can be found for various smart wearable devices that collect some kinds of PPG signals. In addition, since multiple third training samples are further generated by each second indicator prediction model, and the third initial model is trained based on the multiple third training samples to obtain a regression model, a foundation is laid for the subsequent calibration of the results of the second indicator prediction model deployed on the smart wearable device.

[0047] On the basis of the above embodiments, as an optional embodiment, a plurality of second training samples corresponding to the same partial types of PPG signals are combined with the first indicator prediction model to train a second initial model corresponding to the partial types of PPG signals through knowledge distillation to obtain a corresponding second indicator prediction model, including: The second initial model is iteratively trained for multiple rounds in the following manner until the first loss value of the second initial model converges to obtain a corresponding second indicator model: In each round of iterative training, the second training sample is input into the corresponding second initial model to obtain a second physiological indicator interval output by the second initial model; Obtaining a first physiological indicator interval output by the first indicator prediction model for the corresponding first training sample, and combining the second physiological indicator interval to obtain a first sub-value, wherein the first sub-value is used to represent the difference between the first physiological indicator interval and the second physiological indicator interval; wherein the first loss value includes the first sub-value; The model parameters of the second initial model in this iteration are adjusted according to the first loss value.

[0048] See also Figure 4, which exemplarily shows the processing process of one round of iteration of the second initial model of the embodiment of the present application. As shown in the figure, the second training sample is input into the corresponding second initial model to obtain the second physiological indicator interval output by the second initial model. Since the second training sample has a corresponding first training sample, the first training sample can be input into the first indicator prediction model during the first round of iteration of the second initial model to obtain the first physiological indicator interval output by the first indicator prediction model for the first training sample. The first sub-value is determined by calculating the difference between the first physiological indicator interval and the second physiological indicator interval, and the model parameters of the second initial model in this round of iteration are adjusted based on the first sub-value.

[0049] Based on the above embodiments, as an optional embodiment, the first loss value includes a second sub-value, and the second sub-value is used to characterize the difference between the second physiological indicator interval and the sample label.

[0050] In addition to considering the difference in prediction accuracy between the first indicator prediction model and the second initial model, the first loss value of the embodiment of the present application also considers the prediction accuracy of the second initial model for the second training sample, and optimizes the model parameters of the second initial model from two dimensions to obtain a second indicator prediction model with higher accuracy.

[0051] See also Figure 5 , which exemplarily shows a schematic diagram of the training process processing flow of the second initial model in an embodiment of the present application, as shown in the figure, including: S201, obtaining a plurality of second training samples corresponding to the PPG signal types of the second initial model, and a first training sample corresponding to each second training sample; S202, inputting the first training sample corresponding to each second training sample into the first indicator prediction model, and obtaining the first physiological indicator interval corresponding to each first training sample output by the first indicator prediction model; S203, inputting the second training sample into the second initial model of this iteration, and obtaining a second physiological indicator interval output by the second initial model; S204, obtaining a first sub-value according to the first physiological indicator interval and the second physiological indicator interval, and obtaining a second sub-value according to the second physiological indicator interval and the sample label, wherein the first loss value includes the first sub-value and the second sub-value; S205, determining whether the first loss value converges, if not, executing S206, if yes, executing S207; S206, adjusting the model parameters of the second initial model in this iteration according to the first loss value, obtaining the second initial model for the next iteration, and returning to step S203; S207. Use the second initial model of this iteration as the second indicator prediction model.

[0052] Based on the above embodiments, as an optional embodiment, a third initial model is trained with multiple third training samples to obtain a regression model, including: The third initial model is iteratively trained for multiple rounds in the following manner until the loss value of the third initial model converges to obtain the regression model: In each round of iterative training, the third training sample is input into the third initial model to obtain the predicted physiological index value output by the third initial model; Obtaining a second loss value of the third initial model according to the predicted physiological indicator value and the training label, wherein the second loss value is used to represent the difference between the predicted physiological indicator value and the training label; The model parameters of the third initial model in this round of iteration are adjusted according to the second loss value.

[0053] The input of the third initial model of the embodiment of the present application is the third training sample, which is the median value of the physiological indicator interval output by multiple second indicator prediction models according to the corresponding part of the PPG signal in the same time period, that is, each second indicator prediction model outputs its own predicted physiological indicator interval for different PPG signals in the same time period, and the median value of each physiological indicator interval is summarized as the third training sample. The third initial model obtains the predicted physiological indicator value by integrating the median values ​​of all physiological indicator intervals, and further uses the difference between the predicted physiological indicator value and the training label as the second loss value of the third initial model, and adjusts the model parameters of the third initial model according to the second loss value.

[0054] Based on the above embodiments, as an optional embodiment, each second training sample includes one type of PPG signal in the corresponding first training sample. When each second training sample includes only one type of PPG signal, the complexity of the model structure of the second initial model and the order of magnitude of the model parameters will be further reduced, and the second indicator prediction model finally trained will not consume too much energy when deployed on the smart wearable device.

[0055] The number of second indicator prediction models of the present application is consistent with the number of all types of PPG signals.

[0056] Based on the above embodiments, as an optional embodiment, the first training sample also includes an electrocardiogram (ECG) signal of the corresponding time period. When the ECG signal is introduced, the dimension of the information contained in the first training sample is further enriched, and the prediction ability of the first indicator prediction model is also improved accordingly, which indirectly improves the prediction ability of each second indicator prediction model.

[0057] See also Figure 6 , which exemplarily shows a flow chart of a model training method provided by another embodiment of the present application, as shown in the figure, including: using a first training sample 100 including a plurality of PPG signals and ECG signals to train a first initial model 110 to obtain a first indicator prediction model, then using a plurality of second training samples 120 corresponding to the same partial type of PPG signals to train a corresponding second initial model 130, the loss value obtained by the second initial model 130 during the training process includes a first sub-value and a second sub-value, wherein the first sub-value 140 represents the difference between the second physiological indicator interval output by the second initial model 130 and the first physiological indicator interval output by the first indicator detection model 110, and the second sub-value 150 represents the difference between the second physiological indicator interval and the sample label. The intermediate value of the physiological indicator interval output by each second indicator detection model is used as the third training sample of the third initial model 160, the third initial model integrates each intermediate value to obtain a predicted physiological indicator value, and performs a difference calculation between the predicted physiological indicator value and the actual physiological indicator value 170 to obtain a regression model.

[0058] It should be noted that the current wearable device's physiological indicator estimation performance will gradually weaken over time. Taking blood pressure as an example, when calibrating the blood pressure estimation performance of wearable devices, users are required to measure systolic blood pressure (SBP) and diastolic blood pressure (DBP) once, such as using a home-use cuff blood pressure measurement device. Currently, recalibration takes several hours, which is too long, and because the application environment of wearable devices is relatively complex - temperature, wearing pressure, eating mood, etc. will affect the calibration coefficient, so it is inappropriate to simply smooth out the difference between the test results and the true value.

[0059] Therefore, the embodiment of the present application also provides a physiological index detection method, which is applied to a smart wearable device, and the smart wearable device deploys a target index prediction model, such as Figure 7 As shown, including: S301, collecting some types of PPG signals in the first period.

[0060] It should be noted that conventional smart wearable devices can only collect some types of PPG signals.

[0061] S302: If it is determined that the target indicator prediction model has not been calibrated for a preset period of time, all types of PPG signals in the second period are obtained, and all types of PPG signals in the second period are uploaded to the server.

[0062] After collecting some types of PPG signals in the first time period, the embodiment of the present application will determine whether the target indicator prediction model deployed on the smart wearable device has not been calibrated for a preset period of time, and will remind the user to re-acquire all types of PPG signals in the second time period. Specifically, the smart wearable device in the embodiment of the present application sends a voice or screen prompt to the user to re-collect more comprehensive PPG signals. The smart wearable device in the embodiment of the present application can also communicate with other detection devices, and other detection devices can collect other types of PPG signals that the smart wearable device cannot collect, so that the PPG signals collected by other detection devices and the smart wearable device together constitute all types of PPG signals.

[0063] That is to say, the smart wearable device of the embodiment of the present application can obtain the predicted value of the physiological index based on some types of PPG signals collected by itself most of the time, and when calibration is needed, it needs to be connected to other detection equipment. For example, the user can buy a smart watch and a blood pressure monitor, wherein the smart watch can only collect one type of PPG signal, while the blood pressure monitor can collect all types of PPG signals. In daily use, the user only needs to use the smart watch to detect physiological indicators at any time. After using it for a period of time, the user can connect the smart watch and the blood pressure monitor to communicate, and collect all types of PPG signals through the blood pressure monitor and the smart watch at the same time, and calibrate the physiological indicators based on the server.

[0064] The server in the embodiment of the present application obtains a first indicator prediction model, a regression model and multiple second indicator prediction models through the model training method provided by the above-mentioned embodiments. It should be understood that the target indicator prediction model is one of the multiple second indicator prediction models; that is, in the embodiment of the present application, the target indicator prediction model is a second indicator prediction model deployed locally on the smart wearable device by the server.

[0065] S303: Receive and display the predicted value of the first physiological indicator for the first time period returned by the server.

[0066] The first physiological index prediction value of the embodiment of the present application is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

[0067] The server of the embodiment of the present application first selects multiple reference models from the first indicator prediction model and multiple second indicator prediction models deployed in the cloud. In some embodiments, the type of the reference model is only the second indicator prediction model. In other embodiments, the type of the reference model can be the second indicator prediction model or the first indicator prediction model.

[0068] The server determines the PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and then inputs it into the corresponding reference model, so as to obtain the third physiological indicator interval of the second time period output by each reference model. The middle value of each third physiological indicator interval is further input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

[0069] Based on the above embodiments, as an optional embodiment, the physiological index detection method further includes: If it is determined that the duration of the uncalibrated target indicator prediction model does not exceed the preset duration, inputting the partial types of PPG signals into the target indicator prediction model to obtain a fourth physiological indicator interval output by the target indicator prediction model; The middle value of the fourth physiological indicator interval is displayed.

[0070] That is to say, when the target indicator prediction model deployed by the smart wearable device does not need to be calibrated, some types of PPG signals collected by the smart wearable device are input into the target indicator prediction model, the fourth physiological indicator interval output by the target indicator prediction model is obtained, and the middle value of the fourth physiological indicator interval is displayed to the user.

[0071] On the basis of the above-mentioned embodiments, the smart wearable device of the embodiment of the present application not only displays the first physiological indicator prediction value returned by the server, but also further constructs a new training sample based on the first physiological indicator prediction value to train the target indicator prediction model, thereby optimizing the target indicator prediction model and avoiding the decline of the model performance of the target indicator prediction model.

[0072] The present application embodiment provides a model training device, which is applied to a server, such as Figure 8 As shown, the model training device may include: a first training module 801, a sample splitting module 802, a second training module 803 and a third training module 804, wherein: A first training module 801 is used to train a first initial model with a plurality of first training samples to obtain a first indicator prediction model, wherein the first training samples include all types of PPG signals within a time period, and the first training samples also include sample labels, and the sample labels are used to indicate the physiological indicator interval in which the actual physiological indicator value of the corresponding time period is located; A sample splitting module 802 is used to split each first training sample into multiple second training samples, each second training sample includes a part of the types of PPG signals in the corresponding first training sample, and the second training sample also includes a training label and is the same as the training label of the corresponding first training sample; The second training module 803 is used to train a second initial model corresponding to the partial types of PPG signals by combining the first indicator prediction model with a plurality of second training samples corresponding to the partial types of PPG signals to obtain a corresponding second indicator prediction model by a knowledge distillation method; each second indicator prediction model is used to be deployed in a smart wearable device that collects a corresponding type of PPG signal; The third training module 804 is used to train the third initial model with multiple third training samples to obtain a regression model. The third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​in the same time period.

[0073] The device of the embodiment of the present application can execute the model training method provided by the embodiment of the present application, and the implementation principle is similar. The actions performed by each module in the model training device of each embodiment of the present application correspond to the steps in the model training method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0074] The present application embodiment provides a physiological index detection device, which is applied to smart wearable devices, such as Fig. 9 As shown, the physiological index detection device may include: a signal acquisition module 901, a signal acquisition module 902 and an index display module 903, wherein: A signal acquisition module 901 is used to acquire some types of PPG signals in a first period of time; A signal acquisition module 902 is used to acquire all types of PPG signals in a second period if it is determined that the target indicator prediction model has not been calibrated for a preset period of time, and upload all types of PPG signals in the second period to a server, wherein the server obtains a first indicator prediction model, a regression model, and a plurality of second indicator prediction models through training by the model training method provided in the first aspect, and the target indicator prediction model is one of the plurality of second indicator prediction models; An indicator display module 903, used for receiving and displaying the first physiological indicator prediction value of the first time period returned by the server; The first physiological indicator prediction value is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

[0075] In an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory. The processor executes the above-mentioned computer program to implement the steps of the above-mentioned model training method and physiological indicator detection method. Compared with the related art, it can be achieved: firstly, a first indicator prediction model is obtained based on a first training sample including all kinds of PPG signals, and then the first training sample is split to obtain multiple second training samples, each second training sample includes a part of the types of PPG signals, and in combination with the first indicator prediction model, a second initial model corresponding to the part of the types of PPG signals is trained through knowledge distillation. The second indicator prediction model obtained by training based on the second training sample can obtain the first indicator prediction model , and then a model with higher accuracy is predicted based on fewer types of PPG signals. Moreover, this solution does not need to limit the number of all types of PPG signals. On the contrary, the more the number, the higher the accuracy. In addition, the trained second indicator prediction model can theoretically cover all models of smart wearable devices on the market. For various smart wearable devices that collect some types of PPG signals, corresponding deployable second indicator prediction models can be found. In addition, since each second indicator prediction model is further used to generate multiple third training samples, and a third initial model is trained based on the multiple third training samples, a regression model is obtained, which lays a foundation for subsequent calibration of the results of the second indicator prediction model deployed on the smart wearable device.

[0076] In an alternative embodiment, an electronic device is provided, such as Fig.10 As shown, Fig.10 The electronic device 4000 shown includes: a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, such as through a bus 4002. Optionally, the electronic device 4000 may also include a transceiver 4004, which may be used for data interaction between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present application.

[0077] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0078] The bus 4002 may include a path for transmitting information between the above components. The bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 4002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 4002 is represented by only one thick line in the figure, but it does not mean that there is only one bus or one type of bus.

[0079] The memory 4003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0080] The memory 4003 is used to store the computer program for executing the embodiment of the present application, and the execution is controlled by the processor 4001. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the above method embodiment.

[0081] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0082] The embodiment of the present application also provides a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiment when executed by a processor.

[0083] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that shown or described in the drawings.

[0084] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios of execution time, the execution order of these sub-steps or stages may be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0085] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A model training method, characterized in that: Applied to a server, the method comprises: Training a first initial model with a plurality of first training samples to obtain a first indicator prediction model, wherein the first training samples include all types of photoplethysmographic pulse wave (PPG) signals within a time period, and the first training samples also include sample labels, and the sample labels are used to indicate a physiological indicator interval in which an actual physiological indicator value of the corresponding time period is located; Splitting each first training sample into a plurality of second training samples, each second training sample including a portion of the PPG signals in the corresponding first training sample, and the second training sample also including a training label that is the same as the training label of the corresponding first training sample; Using a plurality of second training samples corresponding to the same partial types of PPG signals, combined with the first indicator prediction model, a second initial model corresponding to the partial types of PPG signals is trained through knowledge distillation to obtain a corresponding second indicator prediction model, each second indicator prediction model is used to be deployed in a smart wearable device that collects a corresponding type of PPG signal; The third initial model is trained with multiple third training samples to obtain a regression model, wherein the third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​in the same time period.

2. The method according to claim 1, characterized in that: The method of training a second initial model corresponding to the partial types of PPG signals by using a plurality of second training samples corresponding to the same partial types of PPG signals and combining the first indicator prediction model with the knowledge distillation method to obtain a corresponding second indicator prediction model includes: The second initial model is iteratively trained for multiple rounds in the following manner until the first loss value of the second initial model converges to obtain a corresponding second indicator model: In each round of iterative training, the second training sample is input into the corresponding second initial model to obtain a second physiological indicator interval output by the second initial model; Obtaining a first physiological indicator interval output by the first indicator prediction model for the corresponding first training sample, and combining the second physiological indicator interval to obtain a first sub-value, wherein the first sub-value is used to represent the difference between the first physiological indicator interval and the second physiological indicator interval; wherein the first loss value includes the first sub-value; The model parameters of the second initial model in this iteration are adjusted according to the first loss value.

3. The method according to claim 2, characterized in that The first loss value includes a second sub-value, and the second sub-value is used to characterize the difference between the second physiological indicator interval and the sample label.

4. The method according to any one of claims 1 to 3, characterized in that: The step of training the third initial model with a plurality of third training samples to obtain a regression model comprises: The third initial model is iteratively trained for multiple rounds in the following manner until the loss value of the third initial model converges to obtain the regression model: In each round of iterative training, the third training sample is input into the third initial model to obtain the predicted physiological index value output by the third initial model; Obtaining a second loss value of the third initial model according to the predicted physiological indicator value and the training label, wherein the second loss value is used to represent the difference between the predicted physiological indicator value and the training label; The model parameters of the third initial model in this round of iteration are adjusted according to the second loss value.

5. The method according to claim 1, characterized in that Each second training sample includes a type of PPG signal in the corresponding first training sample, and the number of the second indicator prediction models is consistent with the number of types of all types of PPG signals.

6. The method according to claim 1, characterized in that The first training sample also includes an electrocardiogram signal of a corresponding time period.

7. The method according to claim 1, characterized in that The type of physiological indicator is at least one of diastolic blood pressure, systolic blood pressure, blood oxygen, and blood sugar.

8. A method for detecting physiological indicators, characterized in that: Applied to a smart wearable device, the smart wearable device deploys a target indicator prediction model, and the method includes: Collecting some types of PPG signals in the first period; If it is determined that the target indicator prediction model has not been calibrated for a preset period of time, all types of PPG signals in the second period are obtained, and all types of PPG signals in the second period are uploaded to the server, and the server is trained to obtain a first indicator prediction model, a regression model and a plurality of second indicator prediction models by the model training method according to claims 1-7, and the target indicator prediction model is one of the plurality of second indicator prediction models; receiving and displaying the predicted value of the first physiological indicator in the first time period returned by the server; The first physiological indicator prediction value is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

9. The method according to claim 8, characterized in that Also includes: If it is determined that the duration of the uncalibrated target indicator prediction model does not exceed the preset duration, inputting the partial types of PPG signals into the target indicator prediction model to obtain a fourth physiological indicator interval output by the target indicator prediction model; The middle value of the fourth physiological indicator interval is displayed.

10. The method according to claim 8, characterized in that Also includes: Taking a PPG signal corresponding to the target indicator prediction model among the multiple PPG signals in the first time period as a target PPG signal; Creating a fourth training sample, wherein the fourth training sample includes the target PPG signal of the first time period, and the fourth training sample also includes a sample label, wherein the sample label is a physiological indicator interval where the predicted value of the first physiological indicator is located; The target indicator prediction model is trained according to the fourth training sample to update the model parameters of the target indicator prediction model.

11. The method according to claim 8, characterized in that The multiple reference model sets also include the first indicator prediction model.

12. A model training device, characterized in that: Applied to a server, the device comprises: A first training module, configured to train a first initial model with a plurality of first training samples to obtain a first indicator prediction model, wherein the first training samples include all types of PPG signals within a time period, and the first training samples also include sample labels, wherein the sample labels are used to indicate a physiological indicator interval in which an actual physiological indicator value of a corresponding time period is located; A sample splitting module, used to split each first training sample into multiple second training samples, each second training sample includes a part of the types of PPG signals in the corresponding first training sample, and the second training sample also includes a training label and is the same as the training label of the corresponding first training sample; A second training module is used to train a second initial model corresponding to the partial types of PPG signals by combining a plurality of second training samples corresponding to the same partial types of PPG signals with the first indicator prediction model through knowledge distillation to obtain a corresponding second indicator prediction model; each second indicator prediction model is used to be deployed on a smart wearable device capable of collecting a corresponding type of PPG signal; The third training module is used to train the third initial model with multiple third training samples to obtain a regression model. The third training samples include the middle values ​​of the physiological indicator intervals output by multiple second indicator prediction models according to the corresponding partial types of PPG signals in the same time period, and the training labels of the third training samples are the actual physiological indicator values ​​in the same time period.

13. A physiological index detection device, characterized in that: Applied to a smart wearable device, the smart wearable device deploys a target indicator prediction model, and the method includes: A signal acquisition module, used for acquiring some types of PPG signals in a first period of time; A signal acquisition module, for acquiring all types of PPG signals in a second period if it is determined that the target indicator prediction model has not been calibrated for a preset period of time, and uploading all types of PPG signals in the second period to a server, wherein the server is trained to obtain a first indicator prediction model, a regression model, and a plurality of second indicator prediction models by the model training method according to claims 1-7, and the target indicator prediction model is one of the plurality of second indicator prediction models; An indicator display module, used for receiving and displaying the predicted value of the first physiological indicator of the first time period returned by the server; The first physiological indicator prediction value is obtained by the server in the following manner: Determine a plurality of reference models from the first indicator prediction model and a plurality of second indicator prediction models, wherein the plurality of reference models include a plurality of second indicator prediction models; Determine a PPG signal corresponding to each second reference model from all types of PPG signals in the second time period, and input the PPG signal into each reference model respectively, to obtain a third physiological indicator interval of the second time period output by each reference model; The intermediate values ​​of each third physiological indicator interval in the second time period are input into the regression model to obtain the predicted value of the first physiological indicator output by the regression model.

14. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1 to 11.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images