KR102395173B1 - A wearable apparatus to manage health care for companion animal - Google Patents

Abstract

The present invention is a technology related to a wearable device for health care of a companion animal. According to an embodiment of the present invention, the wearable device for health care of a companion animal comprises: a sensor that acquires activity data of a companion animal by a predetermined time unit; a memory for storing the acquired activity data; and a processor that detects signs of abnormal health of the companion animal or manages a weight of the companion animal from the stored activity data by using a deep learning inference model.

Description

A wearable apparatus to manage health care for companion animals {A wearable apparatus to manage health care for companion animal}

The present invention relates to a wearable device for health management of companion animals.

Companion animals are animals that live together with humans, and typical examples include dogs and cats. Recently, people's awareness of companion animals is improving, and the industry related to companion animals is also growing rapidly. Accordingly, the health status of companion animals is of great interest to people, and technologies for observing the condition of companion animals or managing companion animals by combining IoT technology and the like are increasing.

In particular, in the case of companion animals, unlike humans, special attention is required for health management of companion animals in that they cannot speak.

(Patent Document 01) Korea Patent Publication No. 10-2018-0067107 (published on June 20, 2018)

The present disclosure provides a wearable device for health management of companion animals.

The problem to be solved by the present invention is not limited to the above-mentioned problems, and other problems and advantages of the present invention not mentioned can be understood by the following description, and more clearly understood by the embodiments of the present invention will be In addition, it will be understood that the problems and advantages to be solved by the present invention can be realized by means and combinations thereof indicated in the claims.

A first aspect of the present disclosure provides a wearable device for health management of a companion animal, comprising: a sensor for acquiring activity data of a companion animal in a predetermined time unit; a memory for storing the obtained activity data; and a processor that detects abnormal health signs of the companion animal from the stored activity data using a deep learning inference model or performs weight management of the companion animal.

Here, the deep learning inference model may include at least one of a C-LSTM inference model and a GRU inference model.

The processor infers a daily life pattern of the companion animal from the activity data using the deep learning inference model, and detects abnormal signs of the companion animal based on the daily life pattern or performs weight management of the companion animal can be done

The processor may be configured to include: a number of repetitions of a specific behavior of the companion animal estimated from the activity data, a first activity content of the companion animal by analyzing the aspect of the specific behavior, a first daily life pattern of the companion animal, and the same breed By comparing at least two or more of the second activity contents corresponding to the normal activity range of , an abnormal symptom in the health of the companion animal may be detected.

The processor may be configured to: A second activity amount corresponding to a first activity amount of the companion animal estimated from the activity data, a daily life pattern of the companion animal, and a normal activity range of a normal weight group of the same breed or normal activity of an overweight group By comparing at least two or more of the third activity amounts corresponding to the ranges, the obesity risk of the companion animal may be estimated.

The processor may provide the activity log of the companion animal including at least one of the analysis result of the activity content of the companion animal and the analysis result of the activity amount of the companion animal to the user through the application application.

The processor may provide a warning notification to the user when a health abnormality is detected with respect to the companion animal or when an obesity estimation result of the companion animal exceeds a predetermined range.

In addition, it may further include a communication module supporting at least one communication method of Wi-Fi, Bluetooth, and mobile communication.

The processor may include adaptively selecting at least one communication method from among communication methods supported through the communication module.

In addition, further comprising a GPS sensor for receiving a GPS signal, wherein the processor estimates the current location of the device based on the GPS signal, and the communication method according to a battery management scenario according to the current location and situation of the device may include selecting

A second aspect of the present disclosure may be a computer-readable recording medium in which a program for executing a wearable device for health management of companion animals on a computer is recorded.

It is suitable for companion animals, and it is possible to manage the health of companion animals through wearable devices that record the daily life of companion animals.

By combining IoT technology and deep learning technology, it is possible to more quickly detect health abnormalities that may occur in companion animals. These may be health abnormalities that people may not be aware of in advance, and these problems can be addressed through technological approaches.

By using a deep learning inference model to analyze the companion animal's activity data and share the companion animal's activity log through a wearable device for companion animals, it is possible to increase the understanding of companion animals.

By selecting the LSTM inference model or GRU inference model among the deep learning models, it is possible to learn about the behavioral patterns of companion animals, and by using the information obtained in the previous step as an input for the inference model, companion animal customized anomaly detection can be performed. can

1 and 2 are examples of a wearable device for managing the health of a companion animal and a method of managing the health of a companion animal using the wearable device according to an embodiment of the present invention.
3 is an example of a process of acquiring activity data of a companion animal and estimating a specific behavior of the companion animal from the activity data according to an embodiment of the present invention.
4A to 4D are examples of estimating a specific behavior of a companion animal based on animal activity data according to an embodiment of the present invention. 5 is an example illustrating a frequency with respect to a specific behavior estimated with respect to a companion animal according to an embodiment of the present invention.
6 is a diagram for explaining detecting health abnormalities of a companion animal using a deep learning inference model according to an embodiment of the present invention. 7 is a diagram for explaining estimating the obesity risk of a companion animal using a deep learning inference model according to an embodiment. 8 is a table for explaining a battery management scenario of a wearable device according to an embodiment of the present invention.
9 is an example of a wearable device according to an embodiment of the present invention.

Advantages and features of the present invention, and a method for achieving them will become apparent with reference to the detailed description in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments presented below, it can be implemented in a variety of different forms, and should be understood to include all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. . The embodiments presented below are provided so that the disclosure of the present invention is complete, and to completely inform those of ordinary skill in the art to which the present invention pertains to the scope of the invention. In describing the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as "comprise" or "have" are intended to designate that a feature, number, step, operation, component, part, or a combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Some embodiments of the present disclosure may be represented by functional block configurations and various processing steps. Some or all of these functional blocks may be implemented in various numbers of hardware and/or software configurations that perform specific functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors, or by circuit configurations for a given function. Also, for example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages. The functional blocks may be implemented as an algorithm running on one or more processors. Additionally, the present disclosure may employ prior art techniques for electronic configuration, signal processing, and/or data processing, etc. Terms such as "mechanism", "element", "means" and "configuration" will be used broadly. and is not limited to mechanical and physical configurations.

In addition, the connecting lines or connecting members between the components shown in the drawings only exemplify functional connections and/or physical or circuit connections. In an actual device, a connection between components may be represented by various functional connections, physical connections, or circuit connections that are replaceable or added.

Hereinafter, the present disclosure will be described in detail with reference to the accompanying drawings.

1 and 2 are examples of a wearable device 10 for managing the health of a companion animal and a method of managing the health of a companion animal using the wearable device 10 according to an embodiment of the present invention.

Here, the device 10 may be in the form of a wearable device that can be attached to a companion animal's collar, necklace, clothes, bag, or the like. In this case, the device 10 may be manufactured in a size that does not interfere with the activity of the companion animal, and may be attached to the body or clothes of the companion animal.

Referring to FIG. 1 , a device 10 may include a sensor 12 , a memory 13 , a processor 14 , and a communication module 15 .

The sensor 12 may acquire activity data of the companion animal in a predetermined time unit. Here, the sensor 12 may include an acceleration sensor, a gyro sensor, or a GPS sensor.

As an example, the acceleration sensor is a sensor that detects a change in speed, and may detect size information measured from three vectors having x, y, and z axes. For example, when the sensor 12 includes an acceleration sensor, the sensor 12 may collect raw-data corresponding to the acceleration information 10 to 100 times per second. More preferably, the sensor 12 may collect raw-data 30 to 70 times per second. More preferably, the sensor 12 may collect raw-data 50 times per second. Compared to the prior art, the sensor 12 acquires raw-data more frequently per predetermined unit of time (eg, 1 second). Accordingly, the device 10 or the server can infer the daily life pattern of the companion animal more accurately than in the prior art.

As another example, the gyro sensor is a sensor that detects angular velocity, and may collect rotation angle information of the companion animal. For example, when the sensor 12 includes a gyro sensor, the sensor 12 may measure data regarding the moving direction of the companion animal from the angular velocity in units of seconds.

Meanwhile, the sensor 12 may collect activity data of the companion animal at predetermined time units. For example, the sensor 12 may include an acceleration sensor or a gyro sensor, a combination thereof, or other types of sensors.

The GPS sensor may receive location information indicating the current location of the device 10 from a satellite. When the companion animal equipped with the device 10 performs an external activity such as a walk, the location of the companion animal may be tracked through a GPS signal, and the device 10 may provide location information of the companion animal to the user. On the other hand, indoors where the GPS reception signal is weak, the location of the companion animal can be collected through another positioning method using the communication module.

The memory 13 may store the acquired activity data. The companion animal's activity data collected from the sensor may be accumulated through the memory 13 . In addition, the memory 13 may record activity data through the processor 14 and a specific behavior or daily life pattern of the companion animal estimated through the deep learning inference model. In addition, the memory 13 may record the activity log of the companion animal by processing information collected about the daily life pattern or specific behavior of the companion animal through the processor 14 .

The processor 14 may detect health abnormalities of the companion animal from the stored activity data or perform weight management of the companion animal by using a deep learning inference model.

Here, there may be several types of deep learning inference models. For example, the deep learning inference model may include a recurrent neural network (RNN), a convolutional long short-term memory (C-LSTM), a gated recurrent unit (GRU), etc., which are models that deal with sequential data with respect to time.

In order for the inference model of the present invention to detect daily life patterns and health abnormalities of companion animals from activity data of companion animals or perform weight management of companion animals, an RNN structure in which the previous time step is input as the state in the next layer It may be desirable to select a deep learning model of However, although the deep learning model of the RNN structure has several advantages, some problems may occur when long-term log records are input to the corresponding inference model. To solve this problem, the present invention can mainly perform abnormal symptom detection of companion animals using C-LSTM or GRU or an inference model combining them.

For example, selecting an optimal inference model among deep learning inference models stabilizes the performance of the device 10 and enables better inference results to be obtained. In addition, the C-LSTM or GRU model shows better efficiency in terms of gradient loss, runaway, and long term dependency compared to the RNN model.

According to an embodiment of the present invention, the device 10 may select a C-LTSM model and/or a GRU model as a deep learning inference model. The C-LSTM inference model can perform inference dependent on a predetermined time by adding the cell state of the previous layer as a hidden state vector in the time step of each layer. In addition, the device 10 may erase some of the memories up to the previous cell by adding a forget gate in some layers in the C-LSTM inference model. For this reason, the apparatus 10 can prevent the runaway of a function according to an increase in the amount of data to be processed when using the C-LSTM inference model, and can serve to contain meaningful information.

On the other hand, the GRU model has a more standardized structure than the C-LTSM model, and is a model that can be calculated quickly. Unlike the LSTM inference model that uses a cell state vector and a hidden state vector, only a hidden state vector can be used.

According to an embodiment, the processor 14 may infer the daily life pattern of the companion animal from the activity data using the deep learning inference model ( 210 ). On the other hand, the deep learning inference model used in the present invention may be referred to as "NeuroSense Algorithm" below.

For example, the processor 14 may learn the activity data accumulation record for a predetermined long period of time collected from the sensor 12 as learning data. At this time, the activity data may have a graph form in which the magnitude of the speed change is recorded with the passage of time, and the processor 14 stores the activity data in units of time or per day, and is used to infer daily life patterns. data can be processed.

The processor 14 may output an intermediate estimation result obtained by analyzing the number of repetitions of the specific behavior of the companion animal and the aspect of the specific behavior through the neurosense algorithm as the first activity contents of the companion animal. As an example, the processor 14 may infer the daily life pattern of the companion animal that is repeated for a predetermined time based on the first activity content through the neurosense algorithm, and output it as an intermediate estimation result.

Also, the processor 14 may detect an abnormal symptom of the companion animal based on the daily life pattern ( 220 ).

For example, the processor 14 may determine a predetermined action corresponding thereto based on activity data of the companion animal using a neurosense algorithm. Also, the processor 14 may identify a specific behavior related to a health condition from among the determined predetermined behaviors. In addition, the processor 14 may be configured to: the number of repetitions of the specific behavior of the companion animal estimated from the activity data, the first activity contents of the companion animal by analyzing the aspect of the specific behavior, the first daily life pattern of the companion animal, and the same breed By comparing at least two or more of the second activity contents corresponding to the normal activity range of , an abnormal symptom regarding the health of the companion animal may be detected.

As an example, the processor 14 refers to the companion animal's daily life pattern, and when the companion animal performs a specific scratching action more than a threshold value more than usual, it detects an abnormal symptom related to the companion animal's health (eg, skin disease). suspicious signs).

As another example, the processor 14 may detect as an abnormal symptom regarding the health of the companion animal, if there is a lot of toss and downs in sleep behavior or frequently wakes up without a deep sleep by referring to the daily life pattern of the companion animal. In this case, the device 10 may notify the user that an “abnormal symptom” has been found even if it cannot accurately infer the reason for the change in the behavioral pattern of the companion animal. When the detection of such "abnormal signs" is continuously detected for a certain period of time, the device 10 may send a countermeasure or recommendations to the user together.

On the other hand, the scratching or licking behavior of the companion animal may not be unusual even compared with the daily life pattern of the companion animal or the daily life pattern of other companion animals. The Neurosense algorithm may further consider the content of the second activity corresponding to the normal activity range of the same breed in order to make an inference result more suitable for the companion animal.

For example, the Neurosense algorithm may use a database about activity contents corresponding to the normal activity range for each breed and age of the companion animal. It can be noted that more activity data may be collected when the companion animal is younger, and less activity data may be collected when the companion animal is older. For example, the Neurosense algorithm can be applied as an additive or subtraction factor to set a threshold for classifying the excess or deficiency of a specific behavior of the companion animal, such as the age of the companion animal.

According to another embodiment, the processor 14 may perform weight management of the companion animal based on the daily life pattern ( 220 ). The processor 14 may receive activity data, type, age, and weight information as companion animal information to perform weight management of the companion animal. In this case, the device 10 may estimate the obesity risk of the companion animal even if there is no part of the companion animal information described above. For example, even if weight information about the companion animal is not input by the user, the device 10 may estimate the obesity risk of the companion animal based on the age and breed of the companion animal.

Here, the processor 14 may use the GRU as an inference model for the neurosense algorithm. For example, the processor 14 learns in advance the second normal activity range corresponding to the normal weight group of the same breed as the companion animal or the third normal activity range corresponding to the overweight group of the same breed, and uses it It is possible to judge the appropriateness of the weight of a specific companion animal or to estimate the obesity risk of a companion animal.

As an example, the processor 14 calculates the difference between the accumulated activity data of the companion animal and the normal activity amount of the normal/overweight group within the same breed, and reflects this as a factor in the GRU Cell to estimate the obesity risk of the companion animal. can A detailed description thereof will be described later with reference to FIG. 5 .

On the other hand, the processor 14 uploads the intermediate estimation result or output data using the neurosense algorithm to the server, so that the content can be shared with the user. Here, the server may be a server that drives an application that is interlocked with the wearable device 10, and the application provides the user with the activity content analysis result, the activity amount analysis result, or daily life pattern information of the companion animal to the user. can In this case, the application may process the data output through the deep learning inference model in the form of an activity log of the companion animal and provide it to the user.

For example, the user may obtain information about the companion animal's activity data, activity log, daily life pattern, proper weight, obesity risk estimate, and health abnormality from the device 10 through the application application. Also, if a health abnormality is detected for the companion animal, the device 10 may further provide a warning notification to the user through the application application.

The communication module 15 supports at least one communication method among Wi-Fi, Bluetooth, and mobile communication, and the processor 14 may adaptively select a communication method supported through the communication module according to a situation. Here, the mobile communication may include an eMTC communication method supporting LTE.

For example, the processor 14 may estimate the current location of the device 10 based on the received GPS signal, and determine to switch the communication method according to the current location of the device 10 and a battery management scenario according to the situation. . A detailed description of the battery management scenario will be described later with reference to FIG. 8 .

3 is an example of a process of acquiring activity data of a companion animal and estimating a specific behavior of the companion animal therefrom according to an embodiment of the present invention.

First, the device 10 may obtain activity data ( 310 ). Here, by plotting the activity data as size values obtained from various types of sensors, various types of activity data can be compared on the same plane. As an example, the companion animal may acquire x-axis, y-axis, and z-axis activity data from a sensor attached to the device 10 .

Then, the device 10 may store the acquired activity data in a memory ( 320 ).

Next, the device 10 may perform behavior estimation of the companion animal using the neurosense algorithm ( 330 ). Here, the behavior estimation may include estimating the behavior of the companion animal corresponding to the acquired activity data, and further may further include estimating a specific behavior that may be an indicator of health abnormalities among the behavior of the companion animal (340) ). For example, the device 10 may classify specific behaviors as several indicators for estimating health abnormalities of the companion animal among numerous behaviors of the companion animal.

For example, when the device 10 selects a sleeping behavior, a licking behavior, a scratching behavior, a drinking water behavior, etc. as a specific activity as an indicator for informing the health abnormality of the companion animal, the device 10 determines whether the specific behavior is estimated and the number and mode of occurrence of a specific action may be analyzed.

As an example, the processor 14 may classify the sleeping behavior of the companion animal as restful, slightly disturbed, disrupted, or the like, using the classifier model.

As another example, the processor 14 may count the number of licking actions or the number of scratching actions of the companion animal. For example, the processor 14 may use the classifier model to rank Infrequent, Occasional, Elevated, or Severe according to the number of times the companion animal has licked or scratched during the day. It can be classified as

As another example, the processor 14 compares the water drinking behavior of the companion animal with a predetermined average value using the classifier model, and classifies it as Below Average, Average, Above Average, etc. can

4A to 4D are examples of estimating the behavior of a companion animal based on activity data of the companion animal according to an embodiment of the present invention.

According to an example, the activity data acquired through the device 10 may be specifically learned according to a behavior pattern of the companion animal. In this case, an image captured by the companion animal may be used to match the specific behavior of the companion animal with the pattern of the activity data. As an example, a specific behavior of the companion animal is recognized from an image taken of the companion animal, and the specific behavior of the companion animal and activity data of the companion animal matching it are acquired through time synchronization with the activity data obtained from the device 10 . can do. Referring to FIG. 4A , the device 10 overlaps the activity data of the x-axis, y-axis, and z-axis collected through a plurality of sensors, and the pattern indicated by each graph and the behavior of the companion animal through the neurosense algorithm. can match. As an example of FIG. 4A , the companion animal may be estimated to have performed actions such as shaking (Shaking, 401), walking (Walk, 402), Running (403), and shaking (Shaking, 404) for a predetermined period.

Referring to FIG. 4B , since the activity data obtained for each dog may be collected differently depending on the behavioral aspect of a specific dog, the activity data obtained from the dog A and the activity data obtained from the dog B are different even with the same shaking behavior. pattern can be seen.

On the other hand, referring to FIG. 4B , the shaking behavior of dog A (Shaking, 411) and the shaking behavior of dog B (Shaking, 412) may not be accurately distinguished by the human eye, but the neurosense algorithm uses the corresponding companion animal This can be distinguished by learning the behavioral patterns for In this case, the Neurosense algorithm can be used as an additive or subtraction factor to estimate the behavior of each companion animal according to the type and age of the companion animal.

Similarly, referring to FIGS. 4C to 4D , FIG. 4C exemplarily shows activity data regarding the dog A's scratching behavior (Scratching, 431) and the dog B's scratching behavior (Scratching, 432). As another example, FIG. 4D exemplarily shows activity data regarding the drinking behavior and eating behavior of dog A (Drinking and Eating, 441) and the drinking behavior and eating behavior of dog B (Drinking and Eating, 442).

5 is an exemplary diagram for explaining a frequency of a specific behavior of a companion animal estimated according to an embodiment of the present invention. Referring to FIG. 5 , the daily life pattern of the companion animal may be inferred from the behavior of the companion animal estimated for a predetermined period. For example, referring to FIG. 5 , the corresponding companion animal has a high frequency of behavior related to running. In addition, the behavioral frequency regarding walking of the corresponding companion animal is also at an upper-middle level. Here, the degree of high/low, or high/middle/low level of the behavioral frequency may be presented as a relative value compared with the average behavioral frequency of other companion animals.

If a companion animal has little activity, such as walking or running, or if the companion animal's activity level, which was previously abundant, rapidly decreases, the device 10 may display a specific behavior of the companion animal. It is possible to detect abnormal signs of companion animals from the frequency of their activity.

6 is a diagram for explaining the detection of abnormal health signs of companion animals using a neurosense algorithm according to an embodiment of the present invention.

Referring to FIG. 6 , data handled through the C-LSTM inference model among the neurosense algorithms can be largely described in the learning stage, input stage, intermediate output stage, and output stage as follows.

In the input step, the device 10 may input the activity data of the companion animal to the C-LSTM inference model as the input data 410 . Here, the device 10 uses the C-LSTM inference model to input the input (INPUT) value to each layer of the LSTM-inference model. can be done

In the learning step, the learning data 420 may include the activity data accumulation record and second activity contents corresponding to the normal activity range of the same breed. Here, the activity data accumulation record may be a record in which activity data collected in a predetermined time unit is accumulated. The accumulated record of the companion animal's activity data may be individually input to the hidden layer of the C-LSTM inference model, or may be used to infer the repeated daily life pattern of the companion animal.

As an example, when activity data on the companion animal is collected over several days or months, the processor 14 discovers a pattern regarding the behavior estimation of the companion animal through the C-LSTM inference model, and uses the can be set as the normal activity range of Also, the processor 14 may monitor a change in the daily life pattern of the companion animal through activity data newly added after predetermined learning.

Also, the processor 14 may learn an inference model using, as a parameter, a characteristic value for each behavior corresponding to the normal activity range of the same breed. For example, the processor 14 builds a breed-specific database for each companion animal belonging to the same breed, but collects the companion animal's activity data and the animal's behavior corresponding to the corresponding activity data, and uses a supervised learning method to model the C-LSTM model. We can build a training data-set for learning

Here, when the learning data-set is built, the processor 14 groups the main variable and the dependent variable to calculate the correlation between each variable, and the C-LSTM inference model is used to detect the health abnormality of the companion animal. It can be set as a parameter for

The C-LSTM inference model is a kind of neural network model, and may be used to output an intermediate estimation result 430 and output data 440 through the following steps.

As an example, the intermediate estimation result 430 may be a first daily life pattern of the companion animal, estimation of a specific behavior of the companion animal, the number of repetitions of a specific behavior, and contents of the first activity of the companion animal, such as analysis of patterns, and the like.

For example, the C-LSTM inference model can extract features of N-grams through convolution. As an example, the one-dimensional synthesis product may include a filter vector that slides over the sequence and detects features at different positions. Through this, n feature maps generated for n filters having the same length may be rearranged into feature representations for each window.

In addition, the C-LSTM inference model performs maximum overpooling or dynamic K-max pooling through Max-Pooling, and among the rearranged features through convolutional product, features related to specific behaviors required to detect health abnormalities in companion animals. can be selected.

The C-LSTM inference model may have at least two or more layers, and a cell state in each layer may be input as an input to the next layer. As an example, an intermediate output may be input as a vector in a first layer and a subsequent layer of C-LSTM.

Each layer of C-LSTM receives the previous operation result as input to the next layer, and some layers may include a forget gate to prevent overfitting of the inference model. The FC Layer (Fully connected Layer) may perform linear transformation according to preset parameters and output, as a final result, an abnormal symptom detection result for a companion animal as a probability.

In this case, various types of classifiers (SVMs) may be used in the C-LSTM model, and may include setting a variable to be preferentially applied in each layer and an appropriate weight for each variable.

In the output step, the C-LSTM inference model may detect health abnormalities of the companion animal as the output data 440 through at least two or more LSTM layers and a final layer (FC layer).

Here, when activity data for a companion animal is input, the processor 14 estimates the behavior of the companion animal from the activity data through the C-LSTM inference model, and provides specific behaviors related to health abnormalities among the behaviors of the companion animal. By scoring a score for each behavior and calculating a score for each behavior, it is possible to analyze the behavioral aspect of the behavior of the companion animal. In addition, the analyzed activity history (second activity content) of the companion animal may be compared with a normal activity range by comparing it with a predetermined threshold.

For example, if several behaviors are estimated from activity data in the first stage through the C-LSTM model, specific behaviors and behavior patterns related to health abnormalities are classified among the behaviors estimated in the first stage in the second stage. Thus, the content of the first activity may be analyzed. Then, based on the content of the first activity analyzed in the second step, abnormal health signs of the companion animal may be detected.

7 is a diagram for explaining estimating the obesity risk of a companion animal using a neurosense algorithm according to an embodiment. Here, as an example of the Neurosense algorithm, judgment of the appropriateness of the weight of the companion animal and the estimation of the risk of obesity of the companion animal through the GRU Cell will be described.

Referring to FIG. 7 , an input step, a learning step, and an output step (including an intermediate estimation result) in which each input data is input in parallel using a GRU cell may be described as follows.

In the input step, the processor 14 may obtain the weight and activity data of a specific companion animal, and may input it to the neurosense algorithm as input data 510 . In this case, the input data 510 is companion animal information, and may include activity data, type of companion animal, age, weight, and the like. The processor 14 may input the above information into the neurosense algorithm by combining at least one or a plurality of pieces of information.

The processor 14 may analyze the activity amount of the companion animal from activity data of the companion animal input through the Neurosense algorithm. Here, the activity data may be used as a hidden state vector by inputting a plurality of activity data collected in different time ranges according to a predetermined time unit to each GRU Cell in parallel.

In the learning step, the learning data 520 includes the activity data accumulation record, the second activity amount corresponding to the normal activity range of the normal weight group of the same breed, and/or the third activity amount corresponding to the normal activity range of the overweight group of the same breed. may include

Here, activity data for companion animals of several breeds rather than a specific companion animal may be collected in advance. For example, a database may be constructed in which the activity data collected for the companion animal of the same breed as the companion animal and the corresponding second and/or third activity amount are used as the learning data-set.

In the GRU Cell, each state of input data may be input as a hidden state vector in the next step. Referring to FIG. 5 , the GRU Cell may calculate a final output through a predetermined intermediate estimation step.

Here, the intermediate estimation result 530 is a result estimated in the intermediate process for determining the appropriateness of the companion animal's weight and estimating the obesity risk of the companion animal, and may be made at any stage of the GRU Cell. For example, the intermediate estimation result 530 may include the first activity amount of the companion animal estimated from the activity data and the first daily life pattern of the companion animal.

In the final output step, the processor 14 may output the result of determining the appropriateness of the weight of the companion animal and the result of estimating the risk of obesity of the companion animal as the output data 540 . The obesity risk estimation result may be presented as a probability value or an index corresponding to the probability value.

Here, the processor 14 is configured to use a neurosense algorithm to generate a second companion animal group corresponding to the normal activity range of the normal weight group among other companion animal groups having the same type as the first activity amount of the companion animal estimated from the acquired activity data using the neurosense algorithm. By comparing the activity amount or the third activity amount corresponding to the normal activity range of the overweight group, the appropriateness of the weight of the specific companion animal may be determined, and the obesity risk of the specific companion animal may be estimated.

8 is a table for explaining a battery management scenario of a wearable device according to an embodiment of the present invention. Referring to FIG. 6 , the device 10 may perform battery management by selecting an appropriate communication method according to a situation.

For example, when the companion animal wearing the device 10 is with the guardian, the device 10 preferentially establishes a Bluetooth connection as a communication method with the guardian's terminal (ie, the terminal currently owned by the guardian). You can choose. At this time, when the device 10 communicates with the guardian's terminal only in the Bluetooth method, the lifespan (hereinafter referred to as 'battery life') when the battery of the device 10 is fully charged may be about 4 months, However, the present invention is not limited thereto.

Similarly, when the companion animal wearing the device 10 is at home, Bluetooth and/or eMTC (LTE Cellular) may be applied as a communication mode between the guardian's terminal and the device 10, in this case, every 2 to 5 minutes. As such, communication may be performed between the guardian's terminal and the device 10 . In this case, the battery life of the device 10 may be about 3 months, but is not limited thereto.

In addition, when the companion animal wearing the device 10 is outside, the device 10 may use any one of Bluetooth and eMTC communication methods as a communication method with the guardian's terminal. In this case, communication may be performed between the guardian's terminal and the device 10 in units of 2 to 5 minutes. In this case, the battery life of the device 10 may be about one month, but is not limited thereto.

Meanwhile, when the companion animal disappears, the device 10 may be switched to the rescue mode through the application of the guardian terminal. In this case, the device 10 may receive a GPS signal in units of 1 minute for location tracking based on the GPS signal. Also, at this time, the eMTC method may be used together for communication with the device 10 and the application. In this case, the battery life of the device 10 may be about 2 days, but is not limited thereto.

On the other hand, the communication method described in the above example is according to a plurality of communication modes and battery usage scenarios that can be selected in the device 10, and in some cases, Wi-Fi, Cellular, short-distance wireless communication, etc. Communication methods may be used interchangeably.

9 is an example of a wearable device according to an embodiment of the present invention. The device 10 of the present invention is a kind of smart wearable device that can track the daily life of the companion animal and provide an activity log of the companion animal to the user.

At this time, the data collected and analyzed in the device 10 are directly provided to the user through a communication method such as Bluetooth through an application application interworking with the device 10, or uploaded to a server that manages the application application through Wi-Fi, etc. By doing so, various information can be provided to the user.

Meanwhile, the present invention may be implemented as a computer-readable recording medium in which a program for executing the wearable device 10 for health management of a companion animal is recorded on a computer.

According to an embodiment of the present disclosure, the method according to various embodiments of the present disclosure may be included and provided in a computer program product. Computer program products may be traded between sellers and buyers as commodities. The computer program product is distributed in the form of a machine-readable storage medium (eg compact disc read only memory (CD-ROM)), or via an application store (eg Play Store™) or between two user devices. It can be distributed directly or online (eg, downloaded or uploaded). In the case of online distribution, at least a part of the computer program product may be temporarily stored or temporarily generated in a machine-readable storage medium such as a memory of a server of a manufacturer, a server of an application store, or a relay server.

The steps constituting the method according to the present invention may be performed in an appropriate order unless there is an explicit order or description to the contrary. The present invention is not necessarily limited to the order in which the steps are described. The use of all examples or exemplary terms (eg, etc.) in the present invention is merely for the purpose of describing the present invention in detail, and unless the scope of the present invention is limited by the examples or exemplary terms, unless it is limited by the claims. it is not In addition, those skilled in the art will appreciate that various modifications, combinations, and changes may be made in accordance with design conditions and factors within the scope of the appended claims or their equivalents.

The spirit of the present invention should not be limited to the described embodiments, and not only the claims described below, but also all ranges equivalent to or changed from these claims, fall within the scope of the spirit of the present invention. something to do.

10: Wearable devices for health management of companion animals
12: sensor
13: memory
14: Processor
15: communication module

Claims (10)

In a wearable device for health management of companion animals,
a sensor for acquiring activity data of the companion animal in a predetermined time unit;
a memory for storing the obtained activity data; and
a processor for detecting health abnormalities of the companion animal from the stored activity data using a deep learning inference model or performing weight management of the companion animal;
The processor is
By using the deep learning inference model to learn the accumulated record of activity data of a specific companion animal owned by an individual as learning data, a life pattern unique to the specific companion animal is estimated, and the estimation result is used for normal activity unique to the specific companion animal set the range,
The number of repetitions of the specific behavior of the specific companion animal estimated from the activity data, the content of the first activity of the specific companion animal analyzed by analyzing the pattern of the specific behavior, the first daily life pattern of the specific companion animal, and the same breed An apparatus for detecting an abnormal symptom of the health of the specific companion animal by comparing at least two or more of second activity contents corresponding to a normal activity range of . According to claim 1,
The deep learning inference model comprises at least one of a C-LSTM inference model and a GRU inference model. delete delete The method of claim 1,
The processor is
a first activity amount of the specific companion animal estimated from the activity data;
the daily life pattern of the particular companion animal, and
An apparatus for estimating the obesity risk of the specific companion animal by comparing at least two or more of a second activity amount corresponding to a normal activity range of a normal weight group of the same breed or a third activity quantity corresponding to a normal activity range of an overweight group of the same breed. 6. The method of claim 1 or 5,
The processor is
An apparatus for providing an activity log of the specific companion animal including at least one of an activity content analysis result of the specific companion animal and an activity amount analysis result of the specific companion animal to a user through an application application. 6. The method of claim 1 or 5,
The processor is
If a health abnormality is detected for the specific companion animal,
An apparatus for providing a warning notification to a user when the obesity estimation result of the specific companion animal exceeds a predetermined range. According to claim 1,
Further comprising a communication module supporting at least one communication method of Wi-Fi, Bluetooth and mobile communication,
The processor is
The apparatus comprising adaptively selecting at least one communication method from among communication methods supported through the communication module. 9. The method of claim 8,
Further comprising a GPS sensor for receiving a GPS signal,
The processor is
estimating the current location of the device based on the GPS signal, and selecting the communication method according to a battery management scenario according to the current location and situation of the device. A computer-readable recording medium in which a program for executing the function of the device of claim 1 is recorded on a computer.

Images