Methods, apparatus and equipment for acquiring a heartbeat data pool and training a heartbeat recognition model

Abstract

This application relates to the field of automated heartbeat recognition technology, and provides a method, apparatus, and device for acquiring a heartbeat data pool and training a heartbeat recognition model, avoiding homogenization of training data during the step-by-step optimization of the heartbeat recognition model. In this application, the current heartbeat recognition model obtained from the current training round is acquired; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object; based on the prediction results of the current heartbeat recognition model for each heartbeat data in the heartbeat dataset of the preset object in the current round, the uncertainty of the current round corresponding to each heartbeat data is obtained; based on the uncertainty of the current round corresponding to each heartbeat data, the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition is determined, and the heartbeat data pool for the next round is obtained.

Description

Technical Field

[0001] This application relates to the field of automated heartbeat recognition technology, and in particular to a method and apparatus for acquiring a heartbeat data pool, a method and apparatus for training a heartbeat recognition model, computer equipment, storage medium, and computer program products. Background Technology

[0002] In automated heartbeat recognition technology, the heartbeat type of the examined object can be identified based on an automated heartbeat recognition algorithm. Considering the significant differences between different objects, personalized heartbeat recognition algorithms can be developed for different objects.

[0003] The existing personalized heartbeat recognition algorithms typically involve selecting the subject's personal data and then using that data to train a model, resulting in a personalized heartbeat recognition algorithm for that subject. However, the entire optimization process is a one-time event. If the data selection is inaccurate or the data is highly homogenized, the resulting personalized heartbeat recognition algorithm will have poor recognition performance. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and apparatus for acquiring a heartbeat data pool, a method and apparatus for training a heartbeat recognition model, a computer device, a storage medium, and a computer program product to address the above-mentioned technical problems.

[0005] This application provides a method for obtaining a heartbeat data pool, the method comprising:

[0006] Obtain the current heartbeat recognition model obtained from the current training round; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object;

[0007] Based on the prediction results of the current round of each heartbeat data in the heartbeat dataset by the current heartbeat recognition model, the uncertainty of the current round corresponding to each heartbeat data is obtained;

[0008] Based on the uncertainty of the current round corresponding to each heartbeat data, determine the heartbeat data in the heartbeat data set that meet the preset uncertainty condition, and obtain the heartbeat data pool for the next round.

[0009] In one embodiment, determining the heartbeat data in the heartbeat dataset that meets a preset uncertainty condition based on the current round uncertainty corresponding to each of the heartbeat data, and obtaining the heartbeat data pool for the next round, includes: determining the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition based on the current round uncertainty corresponding to each of the heartbeat data, and obtaining a candidate pool; obtaining unlabeled heartbeat data and labeled heartbeat data based on whether the heartbeat data in the candidate pool is in the heartbeat data pool for the current round; labeling the unlabeled heartbeat data; and obtaining the heartbeat data pool for the next round based on the labeled unlabeled heartbeat data and the labeled heartbeat data.

[0010] In one embodiment, obtaining the heartbeat data to be labeled and the labeled heartbeat data based on whether the heartbeat data in the candidate pool is in the heartbeat data pool of the current round includes: obtaining the labeled heartbeat data based on the heartbeat data in the candidate pool that is in the heartbeat data pool of the current round; and obtaining the heartbeat data to be labeled based on the heartbeat data in the candidate pool that is not in the heartbeat data pool of the current round.

[0011] In one embodiment, obtaining the uncertainty of the current round corresponding to each heartbeat data based on the current round prediction result of the current heartbeat recognition model for each heartbeat data in the heartbeat dataset includes: for the current round prediction result of each heartbeat data in the heartbeat dataset, obtaining the predicted probability value of the heartbeat data in each category based on the current round prediction result of the heartbeat data; and obtaining the uncertainty of the current round corresponding to the heartbeat data based on the difference between the predicted probability values ​​of the heartbeat data in different categories.

[0012] In one embodiment, obtaining the uncertainty of the current round corresponding to the heartbeat data based on the difference between the predicted probability values ​​of the heartbeat data in different categories includes: obtaining the product result of the heartbeat data in each category based on the product of the predicted probability value of the heartbeat data in the same category and the logarithm of the predicted probability value; and obtaining the uncertainty of the current round corresponding to the heartbeat data based on the sum of the product results of the heartbeat data in each category.

[0013] In one embodiment, the method further includes: obtaining a general heartbeat recognition model for multiple objects; obtaining the target prediction result of each heartbeat data in the heartbeat dataset based on the general heartbeat recognition model, and further obtaining the target uncertainty corresponding to each heartbeat data; determining the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition according to the target uncertainty corresponding to each heartbeat data, and obtaining the heartbeat data pool for the first round.

[0014] This application provides a device for acquiring a heart rate data pool, the device comprising:

[0015] The model acquisition module is used to acquire the current heartbeat recognition model obtained from the current training round; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object;

[0016] The uncertainty processing module is used to obtain the uncertainty of the current round corresponding to each heartbeat data based on the prediction result of the current round of each heartbeat data in the heartbeat dataset by the current heartbeat recognition model.

[0017] The data pool acquisition module is used to determine the heartbeat data in the heartbeat data set that meets the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and to obtain the heartbeat data pool for the next round.

[0018] This application provides a method for training a heartbeat recognition model, the method comprising:

[0019] In each round, determine whether the heartbeat recognition model trained in that round meets the preset training termination condition;

[0020] If the conditions are met, a heartbeat recognition model for the preset object is obtained;

[0021] If the conditions are not met, the next round of heartbeat data pool will be obtained through the steps of the above heartbeat data pool acquisition method, and the next round of model training will be carried out.

[0022] In one embodiment, determining whether the heartbeat recognition model obtained from the training round meets the preset training termination condition includes: determining the heartbeat data that meets the preset uncertainty condition in the heartbeat dataset according to the uncertainty of the round corresponding to each heartbeat data, and obtaining the number of non-compliant heartbeats in the round; and determining whether the heartbeat recognition model obtained from the training round meets the preset training termination condition based on the relative size between the number of non-compliant heartbeats in the round and a set value.

[0023] In one embodiment, determining whether the heartbeat recognition model trained in the current round meets the preset training termination condition based on the relative magnitude between the number of non-compliant heartbeats in the current round and a set value includes: if the number of non-compliant heartbeats in the current round is greater than or equal to the set value, then determining whether the number of non-compliant heartbeats in several consecutive rounds decreases with the increase of rounds; the several consecutive rounds include at least the current round; based on the determination result, determining whether the heartbeat recognition model trained in the current round meets the preset training termination condition.

[0024] In one embodiment, determining whether the heartbeat recognition model obtained in the training round meets the preset training termination condition based on the judgment result includes: if the number of non-compliant heartbeats decreases with the increase of the rounds for several consecutive rounds, then obtaining the downward trend; and determining whether the heartbeat recognition model obtained in the training round meets the preset training termination condition based on the relative magnitude between the significance of the downward trend and the significance threshold.

[0025] This application provides a heartbeat recognition model training device, the device comprising:

[0026] The training end judgment module is used to determine whether the heartbeat recognition model trained in each round meets the preset training end conditions in each round.

[0027] The model acquisition module is used to obtain a heartbeat recognition model for the preset object if the conditions are met.

[0028] The training module is used to obtain the next round of heartbeat data pool and perform the next round of model training if the above-mentioned heartbeat data pool acquisition method is not met.

[0029] This application provides a computer device, including a memory and a processor, wherein the memory stores a computer program and the processor executes the above-described method.

[0030] This application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor using the methods described above.

[0031] This application provides a computer program product having a computer program stored thereon, the computer program being executed by a processor using the above-described method.

[0032] This application trains the heartbeat recognition model in multiple rounds, continuously optimizing it step by step. The data used in each round of training is determined by: obtaining the uncertainty of the heartbeat data based on the prediction results of the heartbeat recognition model from the previous round of training; and selecting data based on this uncertainty. This allows for the selection of more typical heartbeat data, addressing the problem of severe homogenization in training data. Specifically, in each round, based on the heartbeat recognition model trained in that round, the prediction results of each heartbeat data in the heartbeat dataset of the preset object are used to obtain the uncertainty of each heartbeat data in that round. Based on the uncertainty of each heartbeat data in that round, the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition is determined, resulting in the heartbeat data pool for the next round of model training. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating a method for obtaining data from a central data pool in one embodiment.

[0034] Figure 2 This is a schematic diagram of the central image data in one embodiment;

[0035] Figure 3 This is a flowchart illustrating a method for training a center-shot recognition model in one embodiment.

[0036] Figure 4 This is a diagram illustrating the application environment of the method provided in this application in one embodiment;

[0037] Figure 5 This is a flowchart illustrating the training method for a center-beat recognition model in another embodiment;

[0038] Figure 6 This is a structural block diagram of a device for acquiring data from a centrally captured data pool in one embodiment;

[0039] Figure 7 This is a structural block diagram of a centroid recognition model training device in one embodiment;

[0040] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0042] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0043] When establishing a personalized heartbeat recognition algorithm for a specific subject, with the subject's authorization, heartbeat data of the subject can be collected to obtain the subject's heartbeat dataset. The heartbeat data pool mentioned in this application can originate from the subject's heartbeat dataset. In this embodiment, the subject is also referred to as a preset object. The method for obtaining the heartbeat data pool provided in this embodiment can be executed by a computer device, including... Figure 1 The following steps are shown:

[0044] Step S101: Obtain the current heartbeat recognition model obtained from the current round of training.

[0045] The training in the current round is based on the heartbeat data pool of the current round, which can come from the heartbeat dataset of a preset object.

[0046] The acquisition of the heartbeat dataset for the preset object can include: acquiring the electrocardiogram (ECG) signal of the preset object; annotating the ECG signal with R-wave positions based on an R-wave detection algorithm to obtain the positions on the ECG signal corresponding to the R-wave (hereinafter referred to as R-wave positions); obtaining multiple heartbeat data based on the signals located before and after the R-wave positions on the ECG signal; and obtaining the heartbeat dataset of the preset object based on the multiple heartbeat data. The R-wave detection algorithm is, for example, the PT algorithm. Of course, other types of waves, such as Q waves, can also be annotated. This application embodiment does not limit the types of waves annotated or the number of wave types.

[0047] Furthermore, before labeling the ECG signal based on the R-wave detection algorithm, the ECG signal can be preprocessed. Specifically, this can include: standardizing the data format of different ECG signals from a preset object; and then filtering the ECG signal after standardization. Specifically, standardizing the data format can involve converting different ECG signals to the same sampling rate frequency using a cubic spline sampling algorithm, for example, 250Hz. The filtering process can specifically involve using a Butterworth bandpass filter with a frequency ranging from 0.05Hz to 100Hz.

[0048] Furthermore, when obtaining heartbeat data based on signals located before and after the R-wave position on the electrocardiogram (ECG) signal, the specific steps may include: determining a signal position on the ECG signal that is located before the R-wave position and at a first predetermined time interval from the R-wave position, and using this as the first signal position; determining a signal position on the ECG signal that is located after the R-wave position and at a second predetermined time interval from the R-wave position, and using this as the second signal position; extracting the ECG signal corresponding to the heartbeat (referred to as the heartbeat signal) based on the signal between the first and second signal positions, and obtaining the heartbeat data based on the heartbeat signal.

[0049] The first and second preset durations can be the same or different, and their specific values ​​can be set according to actual needs. In some embodiments, the first preset duration can be set to 0.25s, and the second preset duration can be set to 0.45s, resulting in a heartbeat signal as follows: Figure 2 As shown.

[0050] Furthermore, before extracting the heartbeat signal based on the signals located before and after the R wave position on the electrocardiogram (ECG) signal, abnormal markings on the ECG signal can be removed. For example, if there are multiple markings at the same position on the ECG signal, or if some markings contain explanatory information, they can be deleted.

[0051] Furthermore, obtaining heartbeat data from the heartbeat signal can specifically include: extracting features from the heartbeat signal to obtain heartbeat features, and standardizing the heartbeat features to obtain heartbeat data. The extracted features can be the interval information of the R wave, i.e., the distance between the current heartbeat and the previous heartbeat (denoted as RRI_pre) and the distance between the current heartbeat and the next heartbeat (denoted as RRI_aft). The standardization of the heartbeat features can be performed using the following formula. Where x is the heartbeat characteristic, μ is the mean, and σ is the variance.

[0052] When the current round is not the first round, the heartbeat recognition model trained in the previous round can be obtained. The heartbeat recognition model trained in the previous round can be called the previous round heartbeat recognition model. After obtaining the heartbeat data pool for the current round, the previous round heartbeat recognition model is trained to obtain the heartbeat recognition model for the current round. The heartbeat recognition model for the current round can be called the current heartbeat recognition model.

[0053] When the current round is not the first round, the method for obtaining the heartbeat data pool for the current round is as follows: steps S101 to S103.

[0054] When the current round is the first round, a general heartbeat recognition model can be trained based on the heartbeat data pool of the current round. This general heartbeat recognition model is a heartbeat recognition model for multiple objects and has a certain heartbeat type recognition capability. Therefore, this application fine-tunes the general heartbeat recognition model to obtain a personalized heartbeat recognition model for a preset object. This general heartbeat recognition model can be obtained by training the model based on heartbeat data from multiple objects; specifically, the model can be constructed based on a deep learning model of convolutional neural networks, and this application does not impose too many limitations on this. The method of obtaining the heartbeat data pool for the current round will not be described here.

[0055] Step S102: Based on the prediction results of the current round of each heartbeat data in the heartbeat dataset by the current heartbeat recognition model, obtain the uncertainty of the current round corresponding to each heartbeat data.

[0056] After obtaining the current heartbeat recognition model, the category prediction can be performed on each heartbeat data in the heartbeat dataset of the preset object to obtain the prediction result of each heartbeat data. Since the prediction result is output by the current heartbeat recognition model, in order to distinguish it from the prediction results output by the heartbeat recognition model in other rounds, the prediction result can be called the prediction result of the current round.

[0057] After the heartbeat recognition model predicts the category of the heartbeat data, it can obtain prediction results. In these results, if the predicted probability value of the heartbeat data belonging to a certain category differs significantly from the predicted probability values ​​of the heartbeat data belonging to other categories, it indicates that the prediction by the heartbeat recognition model for that heartbeat data is more certain. If the predicted probability values ​​for the heartbeat data belonging to different categories do not differ significantly, it indicates that the prediction by the heartbeat recognition model for that heartbeat data is uncertain. Therefore, after obtaining the prediction results for each heartbeat data point in the current round, the uncertainty of that round can be obtained.

[0058] Step S103: Based on the uncertainty of the current round corresponding to each heartbeat data, determine the heartbeat data in the heartbeat data set that meet the preset uncertainty conditions, and obtain the heartbeat data pool for the next round.

[0059] Before obtaining the heartbeat data pool for the next round, it's possible to first determine whether the current heartbeat recognition model meets the preset training termination condition based on the uncertainty of each heartbeat data point in the current round. If the preset training termination condition is not met, it indicates that the next round of model training needs to be performed, and in this case, the process can directly proceed to step S103 to obtain the heartbeat data pool for the next round. It's understandable that, before obtaining the heartbeat data pool for the next round, it's also possible to proceed directly to step S103 to obtain the heartbeat data pool without determining whether the current heartbeat recognition model meets the preset training termination condition.

[0060] When determining whether the current heartbeat recognition model meets the preset training termination condition, based on the uncertainty of the current round corresponding to each heartbeat data, it is possible to evaluate whether the current heartbeat recognition model has learned well the features used to identify the type of heartbeat data of the preset object. If the above features have been learned well, it can be determined that the current heartbeat recognition model meets the preset training termination condition. If the above features have not been learned well, it can be determined that the current heartbeat recognition model has not met the preset training termination condition.

[0061] For example, the maximum uncertainty in the current round can be determined from the current round uncertainties corresponding to each heartbeat data. Based on the relative magnitude between the maximum uncertainty in the current round and a first uncertainty threshold, it can be determined whether the current heartbeat recognition model meets the preset training termination condition. Specifically, if the maximum uncertainty in the current round is greater than or equal to a preset value, it is determined that the current heartbeat recognition model does not meet the preset training termination condition; if the maximum uncertainty in the current round is less than a preset value, it is determined that the current heartbeat recognition model meets the preset training termination condition.

[0062] After determining that the current heartbeat recognition model does not meet the preset training termination condition, it indicates that the next round of model training is required. At this time, the heartbeat data pool for the next round of training can be obtained, as described in step S103. Since the uncertainty of each round corresponding to the heartbeat data is based on the heartbeat recognition model obtained from the corresponding round of training, the method described in step S103 can change the distribution of training data for each round, avoiding the selection of homogeneous data.

[0063] In this embodiment, the heartbeat data that meets the preset uncertainty condition is determined based on the uncertainty of the current round corresponding to each heartbeat data. This embodiment does not limit the specific implementation of this method. Some possible implementation methods are as follows: the relative magnitude between the uncertainty of the current round corresponding to the heartbeat data and the second uncertainty threshold is used to determine whether the heartbeat data meets the preset uncertainty condition. If the uncertainty of the current round corresponding to the heartbeat data is greater than or equal to the second uncertainty threshold, the heartbeat data can be determined to meet the preset uncertainty condition. If the uncertainty of the current round corresponding to the heartbeat data is less than the second uncertainty threshold, the heartbeat data can be determined not to meet the preset uncertainty condition.

[0064] The magnitudes of the first and second uncertainty thresholds can be set according to actual needs. In some embodiments, the second uncertainty threshold can be set to 0.61.

[0065] Based on the uncertainty of the current round, heartbeat data that meet the preset uncertainty conditions are selected and can be called candidate heartbeat data, thus obtaining the candidate pool. The candidate heartbeat data in the candidate pool are added to the heartbeat data pool of the current round to obtain the heartbeat data pool of the next round.

[0066] Specifically, one approach is to directly add candidate heartbeat data from the candidate pool to the current round's heartbeat data pool, thus obtaining the heartbeat data pool for the next round. Another approach is to sort the candidate heartbeat data in the candidate pool according to their uncertainty in descending order, and add the top M candidate heartbeat data to the current round's heartbeat data pool, thus obtaining the heartbeat data pool for the next round. This method allows for the continuous addition of heartbeat data with higher uncertainty to the existing heartbeat data pool, enabling maximum optimization of the heartbeat recognition model in the next round of training. Here, M can be set relatively small, such as 2 or 3.

[0067] In the above embodiments, in each round, based on the heartbeat recognition model trained in that round, the prediction result of each heartbeat data in the heartbeat dataset of the preset object in that round is obtained to obtain the uncertainty of each heartbeat data in that round; based on the uncertainty of each heartbeat data in that round, it is determined whether the heartbeat recognition model in that round meets the preset training termination condition; if it does not meet the condition, based on the uncertainty of each heartbeat data in that round, the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition is determined to obtain the heartbeat data pool for the next round, and the more typical heartbeat data is selected to avoid the problem of severe homogenization of training data.

[0068] In one embodiment, step S103, which involves determining the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and obtaining the heartbeat data pool for the next round, includes: determining the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and obtaining a candidate pool; obtaining the heartbeat data to be labeled and the labeled heartbeat data based on whether the heartbeat data in the candidate pool is in the heartbeat data pool for the current round; and after labeling the heartbeat data to be labeled, obtaining the heartbeat data pool for the next round based on the labeled heartbeat data to be labeled and the labeled heartbeat data.

[0069] Furthermore, based on whether the heartbeat data in the candidate pool is in the heartbeat data pool of the current round, the heartbeat data to be labeled and the labeled heartbeat data are obtained, including: obtaining the labeled heartbeat data based on the heartbeat data in the candidate pool that is in the heartbeat data pool of the current round; and obtaining the heartbeat data to be labeled based on the heartbeat data in the candidate pool that is not in the heartbeat data pool of the current round.

[0070] Based on the uncertainty of the current round, heartbeat data that meet the preset uncertainty conditions selected from the heartbeat dataset can be called candidate heartbeat data, thus obtaining the candidate pool. Next, it is determined whether the candidate heartbeat data in the candidate pool is within the heartbeat data pool of the current round. If it is, it indicates that the candidate heartbeat data has been categorized, and it is treated as labeled heartbeat data. If it is not, it indicates that the candidate heartbeat data has not yet been categorized, and it is treated as unlabeled heartbeat data. The unlabeled heartbeat data is then categorized. The categories can be determined according to actual needs, such as normal, premature atrial contractions, atrial escape beats, premature ventricular contractions, ventricular escape beats, left bundle branch tissue, right bundle branch block, artifacts, etc. After labeling the unlabeled heartbeat data in the candidate pool, the heartbeat data for the next round is obtained based on the labeled unlabeled heartbeat data and the labeled heartbeat data in the candidate pool.

[0071] Furthermore, the candidate heartbeat data in the candidate pool can be sorted in descending order of uncertainty for the current round to determine the top M candidate heartbeat data. If the top M candidate heartbeat data are already in the current round's heartbeat data pool, they are considered labeled; otherwise, they are considered unlabeled. When the unlabeled heartbeat data is manually labeled, a smaller value for M (e.g., 2 or 3) can reduce the number of heartbeats labeled. After labeling the unlabeled heartbeat data in the candidate pool, the labeled heartbeat data in the candidate pool is combined to obtain the heartbeat data for the next round.

[0072] In one embodiment, step S102, which involves obtaining the current round uncertainty of each heartbeat data point based on the current round prediction result of the current heartbeat recognition model for each heartbeat data point in the heartbeat dataset, includes: for the current round prediction result of each heartbeat data point in the heartbeat dataset, obtaining the predicted probability value of the heartbeat data point in each category based on the current round prediction result of the heartbeat data point; and obtaining the current round uncertainty of the heartbeat data point based on the difference between the predicted probability values ​​of the heartbeat data point in different categories.

[0073] The prediction result for the current round of each heartbeat data can include the probability that the heartbeat data belongs to a category: [P1, P2, ..., P... i ... P n ], where the subscript indicates the category, and P represents the predicted probability value, for example, P i This represents the predicted probability value that the heartbeat data belongs to category i. Categories can include normal, premature atrial contractions, atrial escape beats, premature ventricular contractions, ventricular escape beats, left bundle branch tissue, right bundle branch block, artifacts, etc.

[0074] For example, in a scenario where a heartbeat recognition model is used to identify whether the heartbeat data of a preset object is normal or premature atrial contractions, the prediction result output by the heartbeat recognition model corresponds to two categories, one of which is normal and the other of premature atrial contractions. The heartbeat recognition model can be considered as a binary classification model.

[0075] After the current heartbeat recognition model predicts the category of a specific heartbeat data point, it can output the prediction result for the current round of the prediction. This prediction result can be denoted as [P1, P2], where P1 represents the predicted probability that the heartbeat data is normal, and P2 represents the predicted probability that the heartbeat data is a premature atrial contraction (PAC). Next, the difference between the predicted probability P1 (normal) and P2 (PAC) can be obtained. A larger difference (e.g., P1 = 0.8, P2 = 0.2) indicates higher certainty in the current heartbeat recognition model's category prediction, while a smaller difference (e.g., P1 = 0.55, P2 = 0.45) indicates lower certainty, thus revealing the uncertainty corresponding to the heartbeat data. Since the same heartbeat data can have corresponding prediction results in different rounds, the same heartbeat data has corresponding uncertainties in different rounds. To distinguish between these, the uncertainty obtained from the prediction result of the current round of the heartbeat data is called the current round uncertainty.

[0076] In some implementations, the uncertainty of the current round corresponding to the heartbeat data is obtained based on the magnitude of the difference between the predicted probability values ​​of the heartbeat data in different categories. This can include: calculating the difference between the predicted probability values ​​of the heartbeat data in any pairwise categories. When there are at least three categories, multiple differences can be obtained. For example, when there are three categories, the obtained differences include: P1-P2, P1-P3, and P2-P3. Based on the differences, an average difference can be obtained. Based on the average difference and a pre-built correspondence, the uncertainty of the current round corresponding to the heartbeat data is obtained. This correspondence is the correspondence between the differences and the uncertainty.

[0077] In one embodiment, the uncertainty of the current round corresponding to the heartbeat data is obtained based on the difference between the predicted probability values ​​of the heartbeat data in different categories, including: obtaining the product result of the heartbeat data in each category based on the product of the predicted probability value of the heartbeat data in the same category and the logarithm of the predicted probability value; and obtaining the uncertainty of the current round corresponding to the heartbeat data based on the sum of the product results of the heartbeat data in each category.

[0078] Based on the prediction results of the current round of heartbeat data, the predicted probability values ​​of heartbeat data in different categories are obtained [P1, P2, ..., P...]. i ... P n Afterwards, we can obtain the product of the predicted probability value and the logarithm of the predicted probability value for the heartbeat data in category 1, P1logP1, and the product of the predicted probability value and the logarithm of the predicted probability value for the heartbeat data in category 2, P2logP2. This yields n product results. Then, we calculate the sum of the product results for each category of the heartbeat data. The sum of these products yields the uncertainty of the current round corresponding to the heartbeat data. Specifically, the negative of the product can be used to... As the uncertainty of the current round corresponding to the heartbeat data, the uncertainty obtained at this time is represented by information entropy.

[0079] In one embodiment, this application also provides a method for obtaining the heartbeat data pool for the first round, including: obtaining a general heartbeat recognition model for multiple objects; obtaining the target prediction result of each heartbeat data in the heartbeat dataset based on the general heartbeat recognition model, and further obtaining the target uncertainty corresponding to each heartbeat data; determining the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition according to the target uncertainty corresponding to each heartbeat data, and obtaining the heartbeat data pool for the first round.

[0080] When predicting heartbeat data based on the general heartbeat recognition model, the first round of training of the step-by-step optimization process has not yet begun. In order to distinguish it from the prediction results of different rounds, the prediction result of the heartbeat data output by the general heartbeat recognition model is called the target prediction result.

[0081] After obtaining the target prediction result of the heartbeat data, the uncertainty corresponding to the heartbeat data can be obtained in the manner described in the above embodiment. Similarly, in order to distinguish it from the uncertainty of different rounds, the uncertainty at this time is called the target uncertainty.

[0082] After obtaining the target uncertainty corresponding to the heartbeat data, heartbeat data that meet the uncertainty condition can be determined from the heartbeat dataset of the preset object, as described in the above embodiments, thereby obtaining the heartbeat data pool for the first round. Based on the heartbeat data pool for the first round, the general heartbeat recognition model can be optimized and trained to obtain the heartbeat recognition model for the first round.

[0083] This application also provides a method for training a heartbeat recognition model, which can be executed by a computer device, including... Figure 3 The following steps are shown:

[0084] Step S301: In each round, determine whether the heartbeat recognition model trained in that round meets the preset training termination condition.

[0085] Step S302: If satisfied, a heartbeat recognition model for the preset object is obtained;

[0086] Step S303: If the condition is not met, then the next round of heartbeat data pool is obtained and the next round of model training is performed by following the steps of any embodiment of the above-described method for obtaining the heartbeat data pool.

[0087] In the next round of training, the heartbeat recognition model obtained in the previous round can be trained based on the heartbeat data pool of the next round, thus optimizing the heartbeat recognition model of the previous round and obtaining the heartbeat recognition model of the next round.

[0088] In this embodiment, the heartbeat recognition model is trained in multiple rounds and continuously optimized step by step. The data used in each round of training is: the uncertainty of the heartbeat data is obtained based on the prediction results of the heartbeat recognition model of the previous round of training. The data selected based on this uncertainty can select more typical heartbeat data, thus solving the problem of serious homogenization of training data in the step-by-step optimization process.

[0089] In one embodiment, step S301, determining whether the heartbeat recognition model trained in each round meets the preset training termination condition, may specifically include: in each round, obtaining the heartbeat recognition model trained in that round, obtaining the round uncertainty corresponding to each heartbeat data based on the round prediction result of the heartbeat recognition model for each heartbeat data in the heartbeat dataset; and determining whether the heartbeat recognition model trained in that round meets the preset training termination condition based on the round uncertainty corresponding to each heartbeat data.

[0090] When determining whether the heartbeat recognition model obtained in this training round meets the preset training termination condition, based on the uncertainty of this round corresponding to each heartbeat data, it is possible to evaluate whether the heartbeat recognition model has learned well the features used to identify the type of heartbeat data of the preset object. If the above features have been learned well, it can be determined that the heartbeat recognition model meets the preset training termination condition. If the above features have not been learned well, it can be determined that the heartbeat recognition model has not met the preset training termination condition.

[0091] For example, the maximum uncertainty can be determined from the uncertainty of each round corresponding to each heartbeat data. Based on the relative magnitude between the maximum uncertainty and a first uncertainty threshold, it can be determined whether the heartbeat recognition model meets the preset training termination condition. Specifically, if the maximum uncertainty is greater than or equal to a preset value, it is determined that the heartbeat recognition model does not meet the preset training termination condition; if the maximum uncertainty is less than the preset value, it is determined that the heartbeat recognition model meets the preset training termination condition.

[0092] In one embodiment, determining whether the heartbeat recognition model obtained in the training round meets the preset training termination condition in step S301 includes: determining the heartbeat data that meets the preset uncertainty condition in the heartbeat dataset according to the uncertainty of each heartbeat data in the training round, and obtaining the number of non-compliant heartbeats in the training round; and determining whether the heartbeat recognition model obtained in the training round meets the preset training termination condition based on the relative size between the number of non-compliant heartbeats in the training round and the set value.

[0093] Taking the 8th round in step S301 as an example:

[0094] Based on the uncertainty of the 8th round corresponding to each heartbeat data, heartbeat data that meet the preset uncertainty condition are identified in the heartbeat dataset, and the number of non-compliant heartbeats M_8 in the 8th round is obtained. Based on the relative size between the number of non-compliant heartbeats M_8 and the set value, it is determined whether the heartbeat recognition model trained in the 8th round (denoted as Model_8, and the heartbeat recognition models trained in different rounds mentioned below can be represented in this way) meets the preset training termination condition.

[0095] Specifically, if the number of missed heartbeats M_8 is greater than or equal to the set value, then Model_8's prediction for a large number of heartbeats is uncertain, indicating that Model_8 has not yet learned well the features used to identify the type of heartbeat data for this preset object, and it can be determined that Model_8 does not meet the preset training termination condition. If the number of missed heartbeats M_8 is less than the set value, then Model_8's prediction for a small number of heartbeats is uncertain, indicating that Model_8 has learned well the features used to identify the type of heartbeat data for this preset object, and it can be determined that Model_8 meets the preset training termination condition. The above set value can be set according to actual needs, such as setting the set value to 0.

[0096] In the first round of the step-by-step optimization process, if the model is trained based on a general heartbeat recognition model, since the general heartbeat recognition model is trained based on the heartbeat data of multiple objects, the personalized effect is generally not as expected. In this case, it is not necessary to determine whether the general heartbeat recognition model meets the preset training termination condition. Instead, the general heartbeat recognition model can be trained directly based on the heartbeat data pool of the first round.

[0097] To improve computational resource efficiency, if the number of non-compliant heartbeats M_8 in the 8th round is greater than or equal to a set value, it is not necessary to directly determine whether Model_8 meets the preset training termination condition. Instead, another dimension of information can be used for judgment. Based on whether the relevant conditions are met according to the other dimension of information, it is determined whether Model_8 meets the preset training termination condition. Therefore, in one embodiment, determining whether the heartbeat recognition model trained in that round meets the preset training termination condition based on the relative magnitude between the number of non-compliant heartbeats in that round and the set value includes: if the number of non-compliant heartbeats in that round is greater than or equal to the set value, then determining whether the number of non-compliant heartbeats in several consecutive rounds decreases with the increase of rounds; the several consecutive rounds include at least that round; based on the judgment result, determining whether the heartbeat recognition model trained in that round meets the preset training termination condition.

[0098] Taking the 8th round as an example:

[0099] If the number of non-compliant heartbeats M_8 in the 8th round is greater than or equal to a set value, then the number of non-compliant heartbeats for several consecutive rounds, including the 8th round, can be obtained. For example, the number of non-compliant heartbeats M_7 in the 7th round and the number of non-compliant heartbeats M_8 in the 8th round can be obtained, thus yielding M_7 and M_8. Another example is obtaining the number of non-compliant heartbeats for several consecutive rounds before the 8th round, such as the number of non-compliant heartbeats M_5, M_6, and M_7 in rounds 5 to 7, thus yielding M_5, M_6, M_7, and M_8.

[0100] Given M_7 and M_8, we can determine whether M_7 and M_8 decrease with each round, that is, whether M_8 is less than or equal to M_7.

[0101] If M_7 and M_8 decrease with increasing rounds, it indicates that the uncertainty of Model_8 in predicting the category is lower than that of Model_7. The uncertainty of the heartbeat recognition model in predicting the category decreases with increasing rounds, indicating that Model_8 has room for optimization. Continuing training may achieve better results. Therefore, it can be determined that Model_8 has not met the training termination condition and training should continue.

[0102] If M_7 and M_8 do not decrease with increasing rounds (i.e., M_8 is greater than M_7), it indicates that Model_8's uncertainty in class prediction is higher than Model_7's. The uncertainty of the heartbeat recognition model in class prediction does not decrease with increasing rounds, and Model_8 has no room for optimization. Continuing training may not achieve better results. Therefore, Model_8 meets the preset training termination condition and training is stopped. Compared to Model_8, Model_7 has lower uncertainty in class prediction. Model_7 learns better the features used to identify the type of heartbeat data for the preset object. Therefore, Model_7 can be used as the heartbeat recognition model for the preset object.

[0103] Given M_5, M_6, M_7, and M_8, it can be determined whether M_5, M_6, M_7, and M_8 decrease as the round increases.

[0104] If M_5, M_6, M_7, and M_8 decrease with increasing rounds, then the uncertainty of the heartbeat recognition model in predicting the category decreases with increasing rounds. Model_8 has room for optimization, and continuing training may achieve better results. Therefore, it can be determined that Model_8 has not met the training termination condition, and training should continue.

[0105] If M_5, M_6, M_7, and M_8 do not decrease with increasing rounds, then the uncertainty of the heartbeat recognition model in predicting the category does not decrease with increasing rounds. Model_8 has no room for optimization, and continuing training may not achieve better results. Therefore, Model_8 is determined to meet the preset training termination condition, and training is stopped. Next, among Model_5, Model_6, Model_7, and Model_8, the heartbeat recognition model with the fewest non-compliant heartbeats can be selected as the preset target heartbeat recognition model.

[0106] If the number of non-compliant heartbeats decreases with increasing rounds over several consecutive rounds, the uncertainty of the heartbeat recognition model in predicting the category decreases with increasing rounds, indicating that continued training has a certain optimization effect. To further conserve computational resources, the decision to stop training can be further determined by considering the significance of the optimization effect. Furthermore, based on the judgment result, it is determined whether the heartbeat recognition model trained in this round meets the preset training termination condition, including: if the number of non-compliant heartbeats decreases with increasing rounds over several consecutive rounds, then the decreasing trend is obtained; based on the relative magnitude between the significance of the decreasing trend and the significance threshold, it is determined whether the heartbeat recognition model trained in this round meets the preset training termination condition.

[0107] If the number of unsatisfactory heartbeats decreases with the increase of the number of rounds, the significance of the downward trend can be obtained based on the downward trend of the number of unsatisfactory heartbeats in several consecutive rounds.

[0108] Taking the 8th round as an example:

[0109] Given M_7 and M_8, the difference between M_7 and M_8 can be obtained. The larger the difference, the more obvious the downward trend.

[0110] Given M_5, M_6, M_7, and M_8, the difference in the number of unsatisfactory beats between two adjacent rounds can be obtained, such as the difference between M_5 and M_6, M_6 and M_7, and M_7 and M_8. The magnitude of these differences indicates the significance of the downward trend. For example, these differences can be averaged to obtain the average value; the larger the average value, the more significant the downward trend.

[0111] After obtaining the significance of the downward trend, this significance can be compared with a significance threshold. If the significance is greater than or equal to the significance threshold, it can be determined that Model_8 has a large optimization space, and Model_8 has not met the preset training termination condition, so training continues. If the significance is less than the significance threshold, it can be determined that Model_8 has a small optimization space, and Model_8 meets the preset training termination condition, so training stops.

[0112] To better understand the above methods, the following will combine... Figure 4 and Figure 5 This paper describes an application example of the method described above. Figure 4 This is the overall architecture diagram of this embodiment. This embodiment mainly includes two parts: one part is the acquisition of personal tags, and the other part is the model update process. The model update process is described in detail below. Figure 5 Step S501: Obtain the heartbeat recognition model Model_k. Step S502: Obtain the prediction result of the kth round of heartbeat data. Step S503: Determine whether to continue iterative training. If not, proceed to step S504: Obtain a personalized heartbeat recognition model based on the heartbeat recognition model Model_k. If iterative training continues, proceed to steps S505 and S506: Update the heartbeat recognition model Model_k to obtain the heartbeat recognition model Model_k+1, and then proceed to steps S501 to S503 until a personalized heartbeat recognition model is obtained.

[0113] The specific details are as follows. It should be noted that in this embodiment, the set value for comparison with the number of non-compliant heartbeats is set to 0. The uncertainty of the heartbeat data is used to determine whether the heartbeat data meets the preset uncertainty condition. Specifically, it is determined whether the uncertainty of the heartbeat data is greater than 0.61.

[0114] This embodiment specifically includes the following steps:

[0115] Step S1: Obtain the heart rate dataset of the preset object.

[0116] Step S2: Obtain the heart rate data pool P_1 for the first round.

[0117] like Figure 4 As shown, a general heartbeat recognition model can be loaded from the cloud, and the model can be used to predict the heartbeat data in the heartbeat dataset of the preset object to obtain the target prediction result of the heartbeat data, and further obtain the target uncertainty corresponding to each heartbeat data; heartbeat data with a target uncertainty greater than or equal to 0.61 are added to an empty heartbeat data pool P to obtain the heartbeat data pool P_1 of the first round.

[0118] Step S3: Based on P_1, train the general heartbeat recognition model to obtain the heartbeat recognition model Model_1.

[0119] Step S4: Obtain the uncertainty of the first round of each heartbeat data in the heartbeat dataset.

[0120] Model_1 is used to predict each heartbeat in the heartbeat dataset to obtain the prediction result of the first round of heartbeat data. Based on the prediction result of the first round of heartbeat data, the uncertainty of the first round of heartbeat data is obtained.

[0121] Step S5: Determine whether Model_1 meets the preset training termination conditions, and then determine whether to continue training.

[0122] Based on the heartbeat data with an uncertainty greater than 0.61 in the first round, the number of non-compliant heartbeats in the first round, M_1, is obtained, and it is determined whether M_1 is 0.

[0123] If M_1 is 0, then Model_1 can be determined to meet the preset training termination condition. Model_1 can be used as the heartbeat recognition model for the preset object to obtain a personalized heartbeat recognition model.

[0124] If M_1 is not 0, it can be determined that Model_1 does not meet the preset training termination condition, training continues, and the process proceeds to step S6.

[0125] Step S6: Obtain the heartbeat data pool P_2 for the second round.

[0126] Based on the heartbeat data with an uncertainty greater than 0.61 in the first round, the candidate pool for the first round is obtained. The candidate heartbeat data in the candidate pool for the first round are sorted in descending order of uncertainty in the first round. The top M candidate heartbeat data that are not in P_1 are added to P_1 to obtain the heartbeat data pool P_2 for the second round.

[0127] Step S7: Based on P_2, Model_1 is trained to obtain the heartbeat recognition model Model_2.

[0128] Step S8: Obtain the uncertainty of the second round for each heartbeat data in the heartbeat dataset.

[0129] Model_2 is used to predict each heartbeat in the heartbeat dataset to obtain the prediction result of the second round of heartbeat data. Based on the prediction result of the second round of heartbeat data, the uncertainty of the second round of heartbeat data is obtained.

[0130] Step S9: Determine whether Model_2 meets the preset training termination conditions, and then determine whether to continue training.

[0131] Based on the heartbeat data with an uncertainty greater than 0.61 in the second round, the number of non-compliant heartbeats M_2 in the second round is obtained, and it is determined whether M_2 is 0.

[0132] If M_2 is 0, then Model_2 meets the preset training termination condition. Model_2 is used as the heartbeat recognition model for the preset object to obtain a personalized heartbeat recognition model.

[0133] If M_2 is not 0, then we can further determine whether M_1 and M_2 decrease as the number of rounds increases.

[0134] If M_1 and M_2 do not decrease with the increase of rounds, it can be determined that Model_2 meets the preset training termination condition. Then, Model_1 or Model_2 can be used as the preset heartbeat recognition model to obtain a personalized heartbeat recognition model.

[0135] If M_1 and M_2 decrease with increasing rounds, then it can be further determined whether the significance of the downward trend is higher than the significance threshold.

[0136] If the significance of the downward trend is not higher than the significance threshold, it can be determined that Model_2 meets the preset training termination condition. Then, Model_1 or Model_2 can be used as the heartbeat recognition model of the preset object to obtain a personalized heartbeat recognition model.

[0137] If the significance of the downward trend is higher than the significance threshold, it can be determined that Model_2 does not meet the preset training termination condition, training continues, and the process proceeds to step S10.

[0138] Step S10: Obtain the heartbeat data pool P_3 for the third round.

[0139] Based on the heartbeat data with uncertainty greater than 0.61 in the second round, the candidate pool for the second round is obtained. The candidate heartbeat data in the candidate pool for the second round are sorted in descending order of uncertainty in the second round. The top M candidate heartbeat data that are not in P_2 are added to P_2 to obtain the heartbeat data pool P_3 for the third round.

[0140] Step S11: Based on P_3, Model_2 is trained to obtain the heartbeat recognition model Model_3.

[0141] After obtaining the data, continue processing according to the steps described above until a personalized heartbeat recognition model for the preset object is obtained.

[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0143] In one embodiment, such as Figure 6 As shown, a device for acquiring a heartbeat data pool is provided, comprising:

[0144] The model acquisition module 601 is used to acquire the current heartbeat recognition model obtained from the current round of training; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object;

[0145] Uncertainty processing module 602 is used to obtain the current round uncertainty corresponding to each heartbeat data based on the current round prediction result of each heartbeat data in the heartbeat dataset by the current heartbeat recognition model;

[0146] The data pool acquisition module 603 is used to determine the heartbeat data in the heartbeat data set that meets the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and to obtain the heartbeat data pool for the next round.

[0147] In one embodiment, the data pool acquisition module 603 is further configured to: determine the heartbeat data in the heartbeat data set that meet the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and obtain a candidate pool; obtain the heartbeat data to be labeled and the labeled heartbeat data based on whether the heartbeat data in the candidate pool is in the heartbeat data pool of the current round; label the heartbeat data to be labeled, and obtain the heartbeat data pool of the next round based on the labeled heartbeat data to be labeled and the labeled heartbeat data.

[0148] In one embodiment, the data pool acquisition module 603 is further configured to obtain the labeled heartbeat data based on the heartbeat data in the candidate pool that is in the heartbeat data pool of the current round; and to obtain the heartbeat data to be labeled based on the heartbeat data in the candidate pool that is not in the heartbeat data pool of the current round.

[0149] In one embodiment, the uncertainty processing module 602 is further configured to, for the current round prediction result of each heartbeat data in the heartbeat data set, obtain the prediction probability value of the heartbeat data in each category based on the current round prediction result of the heartbeat data; and obtain the current round uncertainty of the heartbeat data based on the difference between the prediction probability values ​​of the heartbeat data in different categories.

[0150] In one embodiment, the uncertainty processing module 602 is further configured to obtain the product result of the heartbeat data in each category based on the product between the predicted probability value and the logarithm of the predicted probability value in the same category; and to obtain the uncertainty of the current round corresponding to the heartbeat data based on the sum of the product results of the heartbeat data in each category.

[0151] In one embodiment, the data pool acquisition module 603 is further configured to acquire a general heartbeat recognition model for multiple objects; obtain the target prediction result of each heartbeat data in the heartbeat dataset based on the general heartbeat recognition model, and further obtain the target uncertainty corresponding to each heartbeat data; determine the heartbeat data in the heartbeat dataset that meets the preset uncertainty condition according to the target uncertainty corresponding to each heartbeat data, and obtain the heartbeat data pool of the first round.

[0152] Specific limitations regarding the acquisition device for the heartbeat data pool can be found in the limitations on the acquisition method for the heartbeat data pool described above, and will not be repeated here. Each module in the aforementioned heartbeat data pool acquisition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0153] In one embodiment, such as Figure 7 As shown, a heartbeat recognition model training device is provided, comprising:

[0154] The training end judgment module 701 is used to determine whether the heartbeat recognition model trained in each round meets the preset training end conditions in each round.

[0155] The model acquisition module 702 is used to obtain a heartbeat recognition model for the preset object if the conditions are met.

[0156] Training module 703 is used to obtain the next round of heartbeat data pool and perform the next round of model training if the above-mentioned heartbeat data pool acquisition method is not met.

[0157] In one embodiment, the training end judgment module 701 is further configured to determine, based on the uncertainty of the round corresponding to each heartbeat data, heartbeat data in the heartbeat dataset that meets the preset uncertainty condition, and obtain the number of non-compliant heartbeats in the round; and based on the relative size between the number of non-compliant heartbeats in the round and the set value, determine whether the heartbeat recognition model trained in the round meets the preset training end condition.

[0158] In one embodiment, the training end judgment module 701 is further configured to determine whether the number of non-compliant heartbeats in a series of consecutive rounds decreases as the number of rounds increases if the number of non-compliant heartbeats in the round is greater than or equal to the set value; the series of consecutive rounds includes at least the round in question; and based on the judgment result, determine whether the heartbeat recognition model trained in the round meets the preset training end condition.

[0159] In one embodiment, the training end judgment module 701 is further configured to: if the number of non-compliant heartbeats decreases with the increase of the number of rounds, obtain a downward trend; and determine whether the heartbeat recognition model trained in that round meets the preset training end condition based on the relative magnitude between the significance of the downward trend and the significance threshold.

[0160] Specific limitations regarding the heartbeat recognition model training device can be found in the limitations of the heartbeat recognition model training method described above, and will not be repeated here. Each module in the aforementioned heartbeat recognition model training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0161] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 8 As shown. The computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data involved in the above-described method embodiments. The network interface is used for communication with external terminals via a network connection. The computer device also includes input / output interfaces, which are connection circuits for exchanging information between the processor and external devices; they are connected to the processor via a bus and are referred to as I / O interfaces. When the computer program is executed by the processor, it implements a method for acquiring a heartbeat data pool and / or a method for training a heartbeat recognition model.

[0162] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0163] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0164] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the various method embodiments described above.

[0165] In one embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0166] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0167] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0168] The above embodiments are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for acquiring a heartbeat data pool, characterized in that, The method includes: Obtain the current heartbeat recognition model obtained from the current training round; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object; Based on the current round prediction results of the current heartbeat recognition model for each heartbeat data in the heartbeat dataset, the uncertainty of the current round corresponding to each heartbeat data is obtained, including: for the current round prediction results of each heartbeat data in the heartbeat dataset, obtaining the predicted probability value of the heartbeat data in each category based on the current round prediction results of the heartbeat data; and obtaining the uncertainty of the current round corresponding to the heartbeat data based on the difference between the predicted probability values ​​of the heartbeat data in different categories. Based on the uncertainty of the current round corresponding to each heartbeat data, determine the heartbeat data in the heartbeat data set that meet the preset uncertainty condition, and obtain the heartbeat data pool for the next round.

2. The method according to claim 1, characterized in that, The step of determining the heartbeat data in the heartbeat data set that meets the preset uncertainty condition based on the current round uncertainty corresponding to each of the heartbeat data, and obtaining the heartbeat data pool for the next round, includes: Based on the uncertainty of the current round corresponding to each heartbeat data, determine the heartbeat data in the heartbeat dataset that meet the preset uncertainty condition, and obtain the candidate pool; Based on whether the heartbeat data in the candidate pool is in the heartbeat data pool of the current round, the heartbeat data to be labeled and the labeled heartbeat data are obtained; The heartbeat data to be labeled is labeled, and the heartbeat data pool for the next round is obtained based on the labeled heartbeat data to be labeled and the labeled heartbeat data.

3. The method according to claim 2, characterized in that, The step of obtaining the heartbeat data to be labeled and the labeled heartbeat data based on whether the heartbeat data in the candidate pool is in the heartbeat data pool of the current round includes: The labeled heartbeat data is obtained based on the heartbeat data in the heartbeat data pool of the current round in the candidate pool; The heartbeat data to be labeled is obtained from the heartbeat data in the candidate pool that is not in the heartbeat data pool of the current round.

4. The method according to claim 1, characterized in that, The uncertainty of the current round corresponding to the heartbeat data is obtained based on the difference between the predicted probability values ​​of different categories of the heartbeat data, including: Based on the product of the predicted probability value and the logarithm of the predicted probability value in the same category of the heartbeat data, the product result of the heartbeat data in each category is obtained; The uncertainty of the current round corresponding to the heartbeat data is obtained by summing the product results of the heartbeat data in each category.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Obtain a general heartbeat recognition model applicable to multiple objects; Based on the target prediction results of each heartbeat data in the heartbeat dataset using the general heartbeat recognition model, the target uncertainty corresponding to each heartbeat data is obtained. Based on the target uncertainty corresponding to each heartbeat data, determine the heartbeat data in the heartbeat data set that meet the preset uncertainty condition, and obtain the heartbeat data pool for the first round.

6. A method for training a heartbeat recognition model, characterized in that, The method includes: In each round, determine whether the heartbeat recognition model trained in that round meets the preset training termination condition; If the conditions are met, a heartbeat recognition model for the preset object is obtained; If the conditions are not met, the next round of heartbeat data pool is obtained and the next round of model training is performed by following the steps of the method described in any one of claims 1 to 5.

7. The method according to claim 6, characterized in that, The step of determining whether the heartbeat recognition model obtained in this round of training meets the preset training termination condition includes: Based on the uncertainty of each round corresponding to the heartbeat data, determine the heartbeat data that meets the preset uncertainty condition in the heartbeat dataset, and obtain the number of non-compliant heartbeats in that round; Based on the relative size between the number of non-compliant heartbeats in this round and the set value, it is determined whether the heartbeat recognition model trained in this round meets the preset training termination condition.

8. The method according to claim 7, characterized in that, The determination of whether the heartbeat recognition model trained in this round meets the preset training termination condition based on the relative size between the number of non-compliant heartbeats in this round and the set value includes: If the number of non-compliant heartbeats in this round is greater than or equal to the set value, then it is determined whether the number of non-compliant heartbeats in several consecutive rounds decreases as the number of rounds increases; the several consecutive rounds include at least this round. Based on the judgment results, determine whether the heartbeat recognition model obtained in this round of training meets the preset training termination conditions.

9. The method according to claim 8, characterized in that, The step of determining whether the heartbeat recognition model obtained in this round of training meets the preset training termination condition based on the judgment result includes: If the number of unsatisfactory heartbeats decreases with the increase of the number of rounds, then a downward trend is obtained; Based on the relative magnitude between the significance of the downward trend and the significance threshold, it is determined whether the heartbeat recognition model obtained in this round of training meets the preset training termination condition.

10. A device for acquiring a heart rate data pool, characterized in that, The device includes: The model acquisition module is used to acquire the current heartbeat recognition model obtained from the current training round; the training of the current round is based on the heartbeat data pool of the current round; the heartbeat data pool of the current round comes from the heartbeat dataset of a preset object; The uncertainty processing module is used to obtain the uncertainty of the current round corresponding to each heartbeat data based on the current round prediction result of the current heartbeat recognition model for each heartbeat data in the heartbeat dataset, including: for the current round prediction result of each heartbeat data in the heartbeat dataset, obtaining the prediction probability value of the heartbeat data in each category based on the current round prediction result of the heartbeat data; and obtaining the uncertainty of the current round corresponding to the heartbeat data based on the difference between the prediction probability values ​​of the heartbeat data in different categories. The data pool acquisition module is used to determine the heartbeat data in the heartbeat data set that meets the preset uncertainty condition based on the current round uncertainty corresponding to each heartbeat data, and to obtain the heartbeat data pool for the next round.

11. A heartbeat recognition model training device, characterized in that, The device includes: The training end judgment module is used to determine whether the heartbeat recognition model trained in each round meets the preset training end conditions in each round. The model acquisition module is used to obtain a heartbeat recognition model for the preset object if the conditions are met. The training module is configured to, if not satisfied, obtain the next round of heartbeat data pool and perform the next round of model training by following the steps of the method described in any one of claims 1 to 5.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 9.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 9.

Images