KR102239462B1 - Method and system for creating learning data to determine physical stability and wearable device thereof - Google Patents

Abstract

According to an embodiment, a system for generating learning data for determining physical stability includes: a feature extracting unit configured to extract a first feature for determining physical stability and a second feature for a label tag from biometric information; A cluster forming unit configured to form at least one cluster based on the first feature; And a label tagging unit configured to generate an evaluation result for the at least one cluster using the second feature, and tag a label on at least one cluster formed based on the generated evaluation result.

Description

A method of generating learning data, a system, and a wearable device thereof for determining physical stability TECHNICAL FIELD [METHOD AND SYSTEM FOR CREATING LEARNING DATA TO DETERMINE PHYSICAL STABILITY AND WEARABLE DEVICE THEREOF}

Hereinafter, embodiments relate to a method, a system, and a wearable device thereof for generating learning data for determining physical stability.

In order to determine the physical stability of the user, technology for determining the level of physical stability of the user by using biometric information such as sleep information, activity information, and heart rate information collected from the user's wearable device is being developed.

Korean Patent Application Publication No. 10-2018-0006692 (published on January 19, 2018)

An object according to an embodiment is to provide a learning data generation method, a system, and a wearable device for determining physical stability that automatically performs a tagging task required to determine physical stability.

According to an embodiment, a system for generating learning data for determining physical stability includes: a feature extracting unit configured to extract a first feature for determining physical stability and a second feature for a label tag from biometric information; A cluster forming unit configured to form at least one cluster based on the first feature; And a label tagging unit configured to generate an evaluation result for the at least one cluster using the second feature, and tag a label on at least one cluster formed based on the generated evaluation result.

The feature extraction unit may be configured to extract the second feature using daily information and time series information included in the biometric information.

The daily information may include a user identification element, a date of birth, a data collection date, and a resting heart rate.

The time series information may include heart rate per unit time according to the passage of time, number of steps per unit time, and sleep information per unit time.

The cluster forming unit may be configured to generate a data set of a plurality of activity intervals each including sleep information and activity information based on the daily information and the time series information for each of the at least one or more clusters.

The label tag unit may be configured to calculate an activity amount of the at least one or more clusters using a data set of the at least one or more clusters and a spare heart rate, and to tag a label on the at least one or more clusters based on the calculated activity amount. .

When the number of at least one or more clusters is plural, the label tag unit may be configured to compare an activity amount of each of the plurality of clusters and to tag a label having a relatively high level to a cluster having a relatively high activity amount.

The label tag unit may be configured to tag a label on the at least one or more clusters using a data set and an average sleep time of the at least one or more clusters.

When the number of at least one or more clusters is plural, the label tag unit may be configured to compare the average sleep time of each of the plurality of clusters, and to tag a label of a relatively low grade to a cluster having a relatively low average sleep time. have.

According to an exemplary embodiment, a method of generating learning data for determining physical stability includes: extracting a first characteristic for determining physical stability and a second characteristic for a label tag from biometric information; Forming at least one cluster based on the first feature; And generating an evaluation result for the at least one cluster using the second feature, and tagging a label on at least one cluster formed based on the generated evaluation result.

The method of generating learning data for determining physical stability includes: after forming the at least one cluster, generating a data set of a plurality of activity intervals each including sleep information and activity information for each of the at least one or more clusters. And determining a level to tag the at least one cluster based on sleep information and activity information.

A wearable device configured to generate learning data for determining physical stability according to an embodiment includes a sensor configured to detect biometric information including sleep information and activity information of a user, and a first sensor configured to detect physical stability from the biometric information. Extracting a second feature for a feature and a label tag, forming at least one cluster based on the first feature, generating an evaluation result for the at least one cluster using the second feature, and generating It may include a processor configured to tag the label to at least one or more clusters formed based on the evaluation result.

A method, system, and wearable device for generating learning data for determining physical stability according to an embodiment are tasks performed by experts by accurately labeling data used when generating a machine learning model for determining physical stability. It can further reduce cost and time.

The learning data generation method for determining physical stability, the system, and the effects of the wearable device thereof according to an embodiment are not limited to those mentioned above, and other effects that are not mentioned are clearly to a person skilled in the art from the following description. It will be understandable.

1 is a block diagram of a system for generating learning data according to an exemplary embodiment.
2 is a block diagram of a processor of a system for generating learning data according to an exemplary embodiment.
3 is a diagram illustrating a method of extracting features of a processor of a system for generating learning data according to an exemplary embodiment.
4 is a diagram illustrating a method of forming clusters of a processor of a system for generating learning data according to an exemplary embodiment.
5 to 8 are diagrams for explaining a method of generating an evaluation result for a cluster of processors of a learning data generation system according to an exemplary embodiment.

Hereinafter, embodiments will be described in detail through exemplary drawings. In adding reference numerals to elements of each drawing, it should be noted that the same elements are assigned the same numerals as possible, even if they are indicated on different drawings. In addition, in describing the embodiment, if it is determined that a detailed description of a related known configuration or function interferes with the understanding of the embodiment, the detailed description thereof will be omitted.

In addition, in describing the constituent elements of the embodiment, terms such as first, second, A, B, (a) and (b) may be used. These terms are for distinguishing the constituent element from other constituent elements, and the nature, order, or order of the constituent element is not limited by the term. When a component is described as being "connected", "coupled" or "connected" to another component, the component may be directly connected or connected to that other component, but another component between each component It should be understood that may be “connected”, “coupled” or “connected”.

Components included in one embodiment and components including common functions will be described using the same name in other embodiments. Unless otherwise stated, the description of one embodiment may be applied to other embodiments, and a detailed description will be omitted in the overlapping range.

1 is a block diagram of a system for generating learning data according to an exemplary embodiment.

Referring to FIG. 1, the learning data generation system 1 according to an exemplary embodiment uses the biometric information of the object collected through the wearable device 10 to automatically determine the physical stability of the object, which is required for machine learning. Is configured to generate. Here, the object may include living things such as humans and animals. The training data generation system 1 may include a wearable device 10 including a sensor 110 and a processor 120 and a storage unit 20.

The sensor 110 is configured to collect biometric information of an object. Here, the biometric information of the object may include sleep information, activity information, heart rate information, and the like. The sleep information may include a total sleep time, an average sleep time, a duration of a rapid eye movement state, which is a duration for each sleep state, and a duration time of a non-rapid eye movement state. The activity information may include an activity amount. The amount of activity can be defined as a cumulative time based on a heart rate reserve. Here, the spare heart rate refers to a value obtained by subtracting the resting heart rate from the maximum heart rate. For example, the amount of activity may be defined as an activity time in which the heart rate of the object is 40% or more of the spare heart rate. The heart rate information may include an object's resting heart rate, a maximum heart rate, and an average heart rate. In addition, the sensor 110 may collect biometric information of an object as data over time.

The processor 120 is configured to generate learning data based on the biometric information of the object collected by the sensor 110. The processor 120 may extract at least one feature from the biometric information of the object.

The storage unit 20 may store biometric information of an object. In addition, the storage unit 20 may store identification information of an object. The identification information of the object may include a user identification element (ID) of the object, a date of birth, a heart rate at rest, and the like. For example, the storage unit 20 may include a database.

In an alternative embodiment, the learning data generation system 1 may generate learning data using pre-stored biometric information in the storage unit 20 instead of the wearable device 10. In this case, the sensor 110 may be omitted.

In an alternative embodiment not shown, the training data generation system 1 may not include a wearable device 10. In other words, the sensor 110 and the processor 120 may be implemented as separate components. In this case, the biometric information collected by the sensor 110 is transmitted to the processor 120 through a communication unit (not shown), and the processor 120 transmits the biometric information or processed learning data thereof to the storage unit 20. Can be saved.

2 is a block diagram of a processor of a learning data generation system according to an embodiment, FIG. 3 is a diagram for explaining a feature extraction method of a processor of the learning data generation system according to an embodiment, and FIG. 4 is an embodiment A diagram for explaining a cluster formation method of a processor of a learning data generation system according to FIG.

2 to 4, the processor 120 of the system for generating learning data according to an embodiment may include a feature extraction unit 121, a cluster forming unit 122, and a label tag unit 123.

The feature extraction unit 121 may be configured to extract a first feature for determining physical stability and a second feature for a label tag.

The feature extracting unit 121 may extract a first feature for determining physical stability using a technique 1211 well known in the art. For example, the feature extraction unit 121 divides the collected biometric information into an active area and a sleep area based on sleep information, and in order to match the time periods of these areas, the data of some of the active areas and the sleep area Only some sections of data can be considered.

The feature extraction unit 121 may extract a second feature from the biometric information of the object and the identification information of the object by using the daily information and the time series information 1212. Here, the daily information may include a user identification element (ID), a date of birth, a data collection date, and a heart rate at rest. Here, the time series information may include heart rate per unit time according to the passage of time, number of steps per unit time, and sleep information per unit time. For example, the unit time may be 1 minute, but is not limited thereto. The time series information used for extracting the second feature may be generally before the start time of training on the day or before the start of dangerous work based on the employee who leaves work at 9 o'clock and 6 o'clock. This time series information may be divided into a sleep section and an activity section for statistical analysis to be described later.

In addition to extracting the first feature by the feature extracting unit 121, extracting the second feature takes into account the fact that the types of techniques for extracting the first feature are very diverse and each feature is different according to the research direction. . In other words, since it is difficult for the feature extraction unit 121 to automatically generate learning data using only the first feature, it additionally extracts the second feature for the label tag used for classifying the learning data.

The cluster forming part 122 is configured to form at least one cluster based on a first characteristic for determining physical stability. For example, the cluster forming unit 122 may form at least one cluster based on unsupervised learning. Such a method of the cluster forming unit 122 may be used when there is no label in the collected data or the number of data including the label is not sufficient for machine learning.

The cluster forming unit 122 may be configured to generate a data set of a plurality of activity sections each including daily information and sleep information and activity information based on the time series information for each of at least one or more clusters. For example, a data set of a plurality of activity intervals may be implemented in a matrix form. In the horizontal direction of the matrix-type data set, the activity intervals per unit time of a specific date (Rth) are sequentially recorded from the activity interval 1221 at the first time to the activity interval 1222 at the last time, and the vertical direction of the data set is Activity intervals at a specific time over several days can be recorded sequentially. That is, one activity section may include biometric information of a specific date and a specific time. For example, one activity section may include a heart rate, a number of steps, and a sleep time on a specific date and time. In addition, the activity section may coincide with the sleep section of the object.

The label tag unit 123 is configured to tag a label to at least one cluster using a second feature for a label tag. First, the label tag unit 123 may generate an evaluation result for a data set of at least one or more clusters. The evaluation result of the data set of at least one or more clusters may indicate a level of physical stability.

Taking a plurality of clusters as an example, the label tag unit 123 may determine the rank of the cluster with the least sleep amount as a "lower" grade, and tag a label of the "lower" grade to the corresponding data set. In addition, the label tag unit 123 may determine a level of a group having a relatively high activity level among the remaining clusters as a “medium” level, and tag a label having a “medium” level on the corresponding data set. In addition, the label tag unit 123 may automatically determine the grade of the last remaining cluster as the “higher” grade.

The label tag unit 123 may generate an evaluation result for the cluster data set in the order of the above-described examples, and tag a label on the cluster based on the generated evaluation result, but the label tag unit 123 must be You are not bound by this order, and you can create evaluation results or tag labels in a variety of ways and in a variety of orders. In other words, there may be no restrictions on the type and number of labels.

As described above, the learning data for the label-tagged cluster may be stored in a database or the like.

The label tag unit 123 may perform statistical analysis for label tags for each group.

In one embodiment, the label tag unit 123 calculates an activity amount of at least one or more clusters using a data set of at least one or more clusters and a spare heart rate, and tags a label to at least one or more clusters based on the calculated activity amount. Can be configured to The extra heart rate can be defined by the following formula.

<Equation 1>

Here, R _j denotes the j-th data set of the data set R, i denotes the _{i-th activity section in the activity section of the data set R j} (i is the activity section 1221 of the first time and the activity section 1222 of the last time. ) Means a value between).

For example, the label tag unit 123 may calculate the activity amount of a specific cluster by the following equation.

<Equation 2>

Here, c denotes each cluster, n _c denotes the number of data _{in cluster c, and R cj} denotes the j-th data set corresponding to cluster c. Therefore, <Equation 2> refers to the cumulative time of the section where the free heart rate is 40% or more. However, the ratio criterion of the spare heart rate for calculating the cumulative time defining the activity amount of the cluster may be selected as 40% as in the above example, but is not limited thereto. For example, the ratio of the free heart rate may be 0%, 10%, 20%, 30%, 40%, and the like.

The label tag unit 123 may compare the cumulative time of each of the plurality of clusters based on Equation 2, and tag a high-grade label in a cluster having a lower cumulative time than a cluster having a relatively high cumulative time. .

In an embodiment, the label tag unit 123 may be configured to tag labels to at least one or more clusters using data sets and average sleep time of at least one or more clusters. The average sleep time of a specific cluster can be calculated by the following equation.

<Equation 3>

Here, c denotes each cluster, n _c denotes the number of data _{in cluster c, and R cj} denotes the j-th data set corresponding to cluster c. In addition, as described above, the sleeping section of the object coincides with the activity section of the object, and the section size may also be the same.

The label tag unit 123 may compare the average sleep time of each of the plurality of clusters based on Equation 3, and tag a label of a relatively low grade to the cluster having a relatively low average sleep time.

The sleep section of the object may be set in consideration of the purpose of performing sleep evaluation at the same time period and the same time interval for all objects while considering the amount of collected biometric information. For example, the time series information used for feature extraction may include the number of steps, heart rate, and sleep time. To collect time series information for a day, three accesses to the database are required, and time series information over two days is stored. Since collection requires 6 accesses to the database, this can incur a lot of overhead. Therefore, in order to reduce the access frequency per object to 3 times, biometric information on the day before education and training may not be collected.

5 to 8 are diagrams for explaining a method of generating an evaluation result for a cluster of processors of a learning data generation system according to an exemplary embodiment.

5 shows test data of independent samples of clusters 1 and 3, FIG. 6 shows test data of independent samples of clusters 1 and 2, and FIG. 7 shows clusters 2 and 3 It shows the calibration data of the independent surface of. 5 and 6, in "Asleep" representing a sleep state, "Restless" representing a turning state during sleep, and "Nobed" representing a state not sleeping, the second cluster and the third cluster are different from the first cluster. Significant differences from clusters can be identified. Accordingly, the first cluster can be sufficiently classified by the total sleep time, and the first cluster is determined as the lowest grade for physical stability. Meanwhile, referring to FIG. 7, it can be seen that the second cluster and the third cluster show a significant difference in the cumulative time of a specific spare heart rate (HRR). Therefore, since the difference between the second and third clusters is weak in the accumulated time of the spare heart rate of 50% or more, the ratio of the spare heart rate is 40% or more, 30% or more, 20% or more, 10% or more, and It may be one selected from 0% or more.

8 shows the results of forming three clusters through the K-means algorithm for features extracted for each sleep section, and then evaluating the effectiveness of the clusters. As the cluster effectiveness evaluation algorithm, when the evaluation score is 0.5 or more, a silhouette algorithm that evaluates that the cluster is valid may be selected. As can be seen in Figure 8, it can be seen that the rest of the results except for the sleep interval 0 o'clock to 6 o'clock exceed 0.5, which is the criterion for evaluating the effectiveness. Since it can be confirmed that the clustering effectiveness is the highest among the sleep periods from 1 o'clock to 5 o'clock, the sleep period for determining physical stability may be determined between 1 o'clock and 5 o'clock.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description to those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as systems, structures, devices, circuits, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Claims (12)

In the learning data generation system for determining physical stability,
A feature extraction unit configured to extract a first feature for determining physical stability and a second feature for a label tag from the biometric information;
A cluster forming unit configured to form at least one cluster based on the first feature; And
A label tag unit configured to generate an evaluation result for the at least one cluster using the second feature, and to tag a label on at least one cluster formed based on the generated evaluation result;
Learning data generation system comprising a.
The method of claim 1,
The feature extraction unit is a learning data generation system configured to extract the second feature by using daily information and time series information included in the biometric information.
The method of claim 2,
The daily information is a learning data generation system including a user identification element, a date of birth, a data collection date, and a resting heart rate.
The method of claim 2,
The time series information is a learning data generation system including heart rate per unit time, step count per unit time, and sleep information per unit time according to the passage of time.
The method of claim 2,
The cluster forming unit is configured to generate a data set of a plurality of activity intervals each including sleep information and activity information based on the daily information and the time series information for each of the at least one or more clusters.
The method of claim 5,
The label tag unit generates training data configured to calculate the activity amount of the at least one or more clusters using the data set and the spare heart rate of the at least one cluster, and tag a label to the at least one cluster based on the calculated activity amount system.
The method of claim 6,
When the number of at least one or more clusters is plural, the label tag unit is configured to compare an activity amount of each of the plurality of clusters, and to tag a label having a higher grade as the cluster having a higher activity amount.
The method of claim 5,
The learning data generation system, wherein the label tag unit is configured to tag a label on the at least one cluster by using the data set and average sleep time of the at least one cluster.
The method of claim 8,
When the number of at least one or more clusters is plural, the label tagging unit compares the average sleep time of each of the plurality of clusters, and the system having a lower average sleep time is configured to tag a lower grade label as the cluster having a lower average sleep time.
In the learning data generation method performed by a learning data generation system that generates learning data for determining physical stability,
Extracting a first feature for determining physical stability and a second feature for a label tag from the biometric information;
Forming at least one cluster based on the first feature; And
Generating an evaluation result for the at least one cluster using the second feature, and tagging a label on at least one cluster formed based on the generated evaluation result;
Learning data generation method comprising a.
The method of claim 10,
After the step of forming the at least one cluster,
Generating a data set of a plurality of activity intervals each including sleep information and activity information for each of the at least one or more clusters; And
Determining a level to tag the at least one cluster based on sleep information and activity information;
Learning data generation method further comprising.
In the wearable device configured to generate learning data for determining physical stability,
A sensor configured to detect biometric information including sleep information and activity information of a user; And
A first feature for determining physical stability and a second feature for a label tag are extracted from the biometric information, at least one cluster is formed based on the first feature, and the at least one A processor configured to generate an evaluation result for the clusters, and to tag a label on at least one cluster formed based on the generated evaluation result;
Wearable device comprising a.

Images