KR20210118640A - Body fat analyzing system and wearable body fat analyzing device - Google Patents

Abstract

The present invention relates to a body fat analyzing system and a wearable body fat analyzing device. The body fat analyzing system according to an embodiment of the present invention includes a physical quantity detection sensor, a gas sensor array, and an analyzer. The physical quantity detection sensor generates a physical quantity signal by detecting a physical factor. The gas sensor array includes a plurality of gas sensors for detecting a target material in the air, and generates concentration signals of the target material, respectively corresponding to the detection signals based on the detection signals generated from the plurality of gas sensors. The analyzer generates body fat consumption information by analyzing each of the physical quantity signal and the concentration signals through a machine learning model.

Description

Body fat analysis system and wearable body fat analysis device

The present invention relates to diet analysis, and more particularly, to a body fat analysis system and a wearable body fat analysis device.

BACKGROUND ART The development of various technologies, including medical technology, is improving the standard of living of human beings. However, changes in the lifestyle of modern people and wrong eating habits are causing an increase in body fat such as obesity, which can lead to various diseases. In order to lead a healthy life, there is a demand for managing health by simply measuring body fat.

In the past, a user's body fat was measured in various ways. For example, indirect calorimetry is a device that evaluates energy metabolism in the body using oxygen, carbon dioxide, and nitrogen wastes measured from a user's respiration. Such a device is non-invasive and can evaluate the type and type of substrate required for energy metabolism, but may have high cost and space limitations due to the stationary device. For example, an apparatus for measuring body fat by acquiring a magnetic resonance image using MRI has high accuracy, but may have high cost and time consumption due to image acquisition. Accordingly, there is a demand for improving the accuracy of the measured result while measuring and analyzing body fat quickly and conveniently at a desired time and place.

The present invention can provide a body fat analysis system and a wearable body fat analysis device capable of improving the accuracy of an analysis result by using the results measured by various biomarkers while measuring and analyzing body fat quickly and conveniently.

A body fat analysis system according to an embodiment of the present invention includes a physical quantity detection sensor, a gas sensor array, and an analyzer. The physical quantity detection sensor generates a physical quantity signal by detecting a physical factor. The gas sensor array includes a plurality of gas sensors for detecting a target material in the air, and generates concentration signals of the target material respectively corresponding to the detection signals based on the detection signals generated from the plurality of gas sensors. The analyzer analyzes each of the physical quantity signal and the concentration signal through the machine learning model to generate body fat consumption information.

As an example, the physical quantity detection sensor may include a heartbeat sensor that detects a heartbeat of an object to generate a first physical quantity signal, and a momentum sensor that detects an exercise amount of the object and generates a second physical quantity signal. The analyzer may analyze each of the first and second physical quantity signals and the concentration signals through the machine learning model to generate body fat consumption information. For example, the physical quantity detection sensor may include at least one physical sensor detecting a physical factor, and the number of the plurality of gas sensors may be greater than the number of the at least one physical sensor.

For example, the target material may be acetone. For example, each of the plurality of gas sensors may have different reactivity with respect to a target material. In one example, the gas sensor array includes a resistance meter that measures an electrical resistance of each of a plurality of gas sensors by a target material based on the detection signals, and a target detected from the plurality of gas sensors based on the measured electrical resistance. It may further include a converter that calculates the concentration of the substance and generates concentration signals based on the calculated concentrations.

For example, the body fat analysis system may further include a gas sampler that traps at least one material different from the target material and passes the target material through the gas sensor array so that the target material is transmitted to the gas sensor array. As an example, the body fat analysis system may further include a fluid controller for controlling a flow of air delivered to the gas sensor array, an environment controller for controlling an environment of air delivered to the gas sensor array, and an environment sensor for sensing the environment. .

A body fat analysis system according to an embodiment of the present invention includes a physical quantity detection sensor, a gas sensor array, and an analyzer. The physical quantity detection sensor generates a physical quantity signal by detecting a physical factor. The gas sensor array includes a plurality of gas sensors that detect a target material in the air, calculates concentrations of the target material detected from the plurality of gas sensors, and analyzes each of the concentrations through a machine learning model, thereby reacting the target material Generate a signal and a target material non-reactive signal. The analyzer generates body fat consumption information based on the physical quantity signal, the target substance response signal, and the target substance non-reaction signal.

For example, the physical quantity detection sensor may include at least one physical sensor for detecting a physical factor, and the number of the plurality of gas sensors may be greater than the number of the at least one physical sensor.

For example, the target material may be acetone, and a first gas sensor and a second gas sensor among the plurality of gas sensors may have different reactivity to acetone. For example, the electrical resistance of each of the plurality of gas sensors may vary depending on the target material, and the gas sensor array may calculate concentrations of the target material based on the change in electrical resistance.

For example, the analyzer may generate body fat consumption information by applying a weight parameter learned to the physical quantity signal, the target substance response signal, and the target substance non-response signal. For example, the analyzer may generate body fat consumption information by applying a linear regression equation to the physical quantity signal, the target substance response signal, and the target substance non-response signal.

A wearable body fat analyzer according to an embodiment of the present invention includes a plurality of gas sensors, a concentration calculator, and an analyzer. A plurality of gas sensors detect a target substance contained in the user's exhalation. The concentration calculator calculates the concentration of the target material detected from each of the plurality of gas sensors based on the reactivity of each of the plurality of gas sensors to the target material. The analyzer determines the body fat consumption by analyzing the concentrations of the target substances respectively detected from the plurality of gas sensors through the machine learning model.

For example, the electrical resistance of each of the plurality of gas sensors may vary depending on the target material, and the concentration calculator may calculate concentrations of the target material respectively corresponding to the plurality of gas sensors based on the change in electrical resistance. For example, the analyzer may analyze the concentrations of the target substance to determine the start time of body fat consumption. For example, a first gas sensor and a second gas sensor among the plurality of gas sensors may have different reactivity with respect to a target material.

For example, the wearable body fat analyzer may further include a physical quantity detection sensor configured to detect a physical factor to generate a physical quantity signal, and the analyzer may further input the physical quantity signal to the machine learning model to determine body fat consumption. For example, the physical quantity detection sensor may include at least one physical sensor detecting a physical factor, and the number of the plurality of gas sensors may be greater than the number of the at least one physical sensor.

According to an embodiment of the present invention, by performing machine learning-based body fat consumption analysis by further considering the physical factor by the physical quantity detection sensor to the result of the detection of exhaled substances by the gas sensor array, the accuracy of analysis of exhaled substances due to moisture and impurities The reduction can be compensated for.

In addition, according to an embodiment of the present invention, by performing the analysis of the exhaled material by the gas sensor array or the analysis of the exhaled material and the physical factor as a machine learning-based learning model, the accuracy of judgment on body fat consumption or diet progress is improved can be

In addition, according to an embodiment of the present invention, it is possible to quickly and simply monitor body fat consumption or diet progress through a wearable device including a gas sensor array and a physical quantity detection sensor.

1 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention.
FIG. 2 is an exemplary block diagram of the gas sensor array of FIG. 1 ;
FIG. 3 is an exemplary diagram for explaining the operation of the learning model of FIG. 1 .
4 is an exemplary block diagram of the gas sensor array of FIG. 1 ;
FIG. 5 is an exemplary diagram for explaining the operation of the learning model of FIG. 4 .
FIG. 6 is an exemplary view for explaining the operation of the learning model of FIG. 1 to which the gas sensor array of FIG. 4 is applied.
7 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention.
8 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention.
FIG. 9 is an exemplary diagram for explaining the operation of the learning model of FIG. 8 .
10 is an exemplary block diagram of an electronic system to which the body fat analysis system described in FIGS. 1 to 9 is applied.

Hereinafter, embodiments of the present invention will be described clearly and in detail to the extent that those skilled in the art can easily practice the present invention.

1 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention. Referring to FIG. 1 , the body fat analysis system 100 may include a physical quantity detection sensor 110 , a gas sensor array 120 , an analyzer 130 , and a communication module 140 . The body fat analysis system 100 of FIG. 1 is a wearable body fat analysis device capable of detecting a user's physical factor and a target substance in exhalation, and analyzing the factors detected with various biomarkers to provide the user with a body fat consumption or diet progress status. can be implemented as

The physical quantity detection sensor 110 may detect a physical factor of an object such as a user. Here, the physical factor may mean physically measurable elements of the object, such as the movement of the heart (heartbeat), the momentum of the object, and the amount of heat of the object. The physical quantity detection sensor 110 may detect at least one physical factor of the object. In order to improve the analysis accuracy of body fat consumption or diet progress, the physical quantity detection sensor 110 may detect a plurality of physical factors.

As an example, as shown in FIG. 1 , the physical quantity detection sensor 110 may include a heart rate sensor 111 and an exercise quantity sensor 112 . However, the present invention is not limited thereto, and the physical quantity detection sensor 110 may further include various physical sensors capable of detecting physical factors of an object, such as a body temperature detection sensor or a respiration amount sensor. Each of these physical sensors may detect a physical factor of an object to generate a physical quantity signal that is an electrical signal. The physical quantity signal may have an electrical value obtained by converting the detected physical factor.

For example, the heart rate sensor 111 may be configured to detect a difference between a transmitted signal output through an antenna and a received signal reflected through the heart in order to detect the movement of the heart. For example, the heart rate sensor 111 may be configured to receive an electrocardiogram signal through an electrode attached to the user to detect heart rate variability. In addition, the heartbeat sensor 111 may detect the heartbeat of the object in various ways. The heartbeat sensor 111 may generate a heartbeat signal HD, which is an electrical physical quantity signal, based on heartbeat detection.

As an example, the momentum sensor 112 may be configured to detect an amount of motion of an object. As an example, the exercise amount sensor 112 may be configured to detect a movement of an object and calculate a calorie consumption. As an example, the momentum sensor 112 may be configured to detect a change in an electrical biosignal according to an object's motion. In addition, the momentum sensor 112 may detect the momentum of the object in various ways. The momentum sensor 112 may generate a momentum signal CD, which is an electrical physical quantity signal, based on the detection of the momentum.

The gas sensor array 120 may detect a target substance in the exhalation of an object. For example, the target material may be acetone. When the body uses body fat as an energy source, ketone bodies can be formed as a product of fat breakdown. Such a ketone body may include acetone, beta hydroxybutyrate, acetic acetate, and the like, of which acetone gas with high volatility may be discharged to the exhaled air. The gas sensor array 120 may detect acetone in exhaled air to determine body fat decomposition.

The gas sensor array 120 may include a plurality of gas sensors SE for detecting a target material. For example, the plurality of gas sensors SE may be arranged in two dimensions, but is not limited thereto, and may be arranged in a one-dimensional row or column. The number of the plurality of gas sensors SE is not limited, and may be, for example, between 2 and 1000. The number of the plurality of gas sensors SE may be greater than the number of physical sensors included in the physical quantity detection sensor 110 . Since the physical sensors may be provided to improve the accuracy of concentrations calculated based on the plurality of gas sensors SE, the number of the gas sensors SE may be greater than that of the physical sensors. By the plurality of gas sensors SE, the number of samples for detecting acetone in exhaled air increases, and the amount of data for analyzing body fat consumption or diet progress may increase. As a result, analysis accuracy can be increased.

The gas sensor array 120 may generate detection signals based on the target material detected by the plurality of gas sensors SE. Each of the plurality of gas sensors SE may include a material having an electrical resistance that changes according to adsorption with a target material. The gas sensor array 120 may calculate the concentrations of the target material detected from each of the plurality of gas sensors SE by measuring the changing electrical resistance of each of the plurality of gas sensors SE. As a result of the calculation, the gas sensor array 120 may generate the concentration signals SD1 to SDn respectively corresponding to the plurality of gas sensors SE.

Each of the plurality of gas sensors SE may have different reactivity with respect to the detected target material. As a result, each of the concentration signals SD1 to SDn may have different electrical values. A substance that reacts with acetone may also react with other substances (eg sulfur oxides and moisture). Accordingly, by configuring the material or catalyst included in each of the plurality of gas sensors SE differently, erroneous detection and reduction in analysis accuracy due to the reaction of other materials may be improved.

The analyzer 130 receives physical quantity signals (eg, a heartbeat signal HD and an exercise quantity signal CD) from the physical quantity detection sensor 110 , and the concentration signals SD1 to SDn from the gas sensor array 120 . can receive The analyzer 130 may analyze various biomarkers, that is, the physical quantity signals HD and CD and the concentration signals SD1 to SDn, based on the machine learning-based learning model 131 . As a result of the analysis of the learning model 131 , an analysis result indicating a degree of body fat consumption or a degree of diet progress of the user may be calculated. Each component of the analyzer 130 may be implemented in hardware, firmware, software, or a combination thereof.

The learning model 131 may be implemented in the analyzer 130 and learned to non-linearly analyze the user's body fat consumption or diet progress from the physical quantity signals HD and CD and the concentration signals SD1 to SDn. have. The learning model 131 may be integrated and managed within the body fat analysis system 100, but is not limited thereto, and the learning model 131 is implemented in a separate server or storage medium, and is then transferred to the analyzer 130 during analysis. can be loaded. The learning model 131 may be built through an artificial neural network or deep learning machine learning. As an example, the learning model 131 is a convolutional neural network (CNN), a recurrent neural network (RNN), a deep neural network (DNN), a K-Nearest Neighbor (KNN), and decision making. It may be composed of at least one of a decision tree (DT), a multi-layer perceptron (MLP), a partial least square method (PLS), and a support vector machine.

The communication module 140 may be configured to allow the body fat analysis system 100 to communicate with another external electronic device through a wired or wireless network. For example, the communication module 140 may transmit body fat consumption information or diet progress information generated by the analyzer 130 to an external electronic device such as a smartphone, a server, and a computer. For example, the communication module 140 may receive a weight parameter of the learning model 131 according to big data learning from an external electronic device.

FIG. 2 is an exemplary block diagram of the gas sensor array of FIG. 1 ; The gas sensor array 120_1 of FIG. 2 will be understood as an exemplary embodiment of the gas sensor array 120 of FIG. 1 . Referring to FIG. 2 , the gas sensor array 120_1 may include a plurality of gas sensors SE1 , SE2 , ?? and SEn, a resistance meter 121 , and a converter 122 .

Each of the plurality of gas sensors SE1 to SEn may detect a target substance in the exhalation of the object. The target material may be acetone, as described above. Each of the plurality of gas sensors SE1 to SEn may include a material having an electrical resistance that changes according to adsorption with acetone. Such materials include metal oxides, such as SnO ₂ , TiO ₂ , ZnO, WO ₃ , In ₂ O ₃ , Fe ₂ O ₃ , CuO, Co ₃ O ₄ , and ZnCo ₂ O _{4 .} one or a mixture or compound thereof. As an example, these metal oxides may include materials having high specific surface area with porosity of nanodots, nanowires, nanorods, nanoparticles, nanosquare particles, or 3D structures (eg, hollow, MOF, fibers, etc.). can For example, the metal oxide may include particles having a nano-level diameter (eg, 1-500 nm).

Furthermore, at least some of the plurality of gas sensors SE1 to SEn may further include a catalyst material for reaction with the metal oxide. Such a catalyst material may be a noble metal catalyst material, and may include, for example, one or a mixture or compound of _{Pt, Pd, Au, Ru, PtO, PdO, and RuO 2 .}

Each of the plurality of gas sensors SE1 to SEn may have different reactivity with respect to the detected target material. To this end, at least a portion of a material or a catalyst material of each of the plurality of gas sensors SE1 to SEn may be different. For example, when the metal oxide included in the first gas sensor is WO ₃ , WO ₃ may have an electrical resistance that varies depending on hydrogen sulfide as well as acetone. That is, in order to distinguish whether the reaction is caused by acetone or hydrogen sulfide, when the metal oxide included in the second gas sensor is WO ₃ and the catalyst is Pt, the reactivity to acetone is larger than the reactivity to hydrogen sulfide. Therefore, the detection accuracy of the acetone concentration can be improved.

The resistance meter 121 may measure the electrical resistance of each of the plurality of gas sensors SE1 to SEn. The resistance meter 121 may measure electrical resistance that changes in each of the plurality of gas sensors SE1 to SEn according to the reaction with acetone. 2 , the reactivity of each of the plurality of gas sensors SE1 to SEn may be different from each other, and each of the plurality of gas sensors SE1 to SEn resists detection signals according to different reactivity. It can output to the measuring device 121 . The resistance meter 121 may detect the reactivity according to time of each of the plurality of gas sensors SE1 to SEn. Here, the reactivity may be defined as a ratio of the resistance of the gas sensor when measuring the user's exhalation to the resistance of the general air gas sensor. The measured electrical resistance may be output to the converter 122 .

The converter 122 generates concentration signals SD1, SD2, ??, SDn of the plurality of gas sensors SE1 to SEn based on the measured electrical resistance of each of the gas sensors SE1 to SEn. can do. For example, the converter 122 may convert electrical resistance values that are analog electrical signals into concentration signals SD1 to SDn that are digital electrical signals. Referring to the histogram shown in the converter 122, acetone-containing values and acetone-free values for each of the plurality of gas sensors SE1 to SEn are shown. Here, the acetone inclusion value may represent a detection value of acetone in the breathing gas. This detection value may depend on the degree of change in electrical resistance due to the acetone reaction, but may be a value that further includes a reaction by a substance other than acetone. The acetone-free value may represent a non-detectable value of acetone in the breathing gas.

The converter 122 may generate the concentration signals SD1 to SDn of the target material corresponding to the plurality of gas sensors SE1 to SEn, respectively, based on the acetone-containing value and the acetone-free value. Depending on the material included in the gas sensor, the change in electrical resistance may vary between gas sensors. For example, according to the reaction with acetone, electrical resistance may increase or decrease, and the amount of change in electrical resistance may be different for each gas sensor. The converter 122 may generate the concentration signals SD1 to SDn by applying a characteristic value according to the physical properties of each of the plurality of gas sensors SE1 to SEn to the acetone-containing value and the acetone-free value.

FIG. 3 is an exemplary diagram for explaining the operation of the learning model of FIG. 1 . The learning model 131_1 of FIG. 3 will be understood as an exemplary embodiment of the learning model 131 of FIG. 1 . As described with reference to FIGS. 1 and 2 , the learning model 131_1 is based on the physical quantity signals (eg, the heartbeat signal HD and the exercise quantity signal CD) and the concentration signals SD1 to SDn. The degree of body fat consumption or the degree of diet progress can be analyzed.

The learning model 131_1 may be implemented as a machine learning-based neural network. The learning model 131_1 may be implemented as an input layer IL, a hidden layer HL, and an output layer OL. It will be understood that these configurations are exemplary, and the learning model 131_1 may be implemented in a layer structure different from that of FIG. 3 according to the type or shape of the neural network. The input layer IL may receive the physical quantity signals HD and CD and the density signals SD1 to SDn. The received physical quantity signals HD and CD and density signals SD1 to SDn may be propagated to the hidden layer HL. According to learning, a weight parameter may be applied to each signal propagated to the hidden layer HL. The learning model 131_1 may calculate a relationship between each signal based on the weight parameter. The output layer OL may output the analysis result of the degree of body fat consumption or the degree of diet progress based on the analysis result propagated from the hidden layer HL.

The analysis result of the learning model 131_1 shows consumption information DO indicating the extent to which the user's body fat is consumed (the extent to which the diet has progressed) and the non-consumption information DO indicating the extent to which the user's body fat is not consumed (the extent to which the diet is not progressed). Information DX may be included. Furthermore, the learning model 131_1 may further calculate a time when body fat consumption starts based on the analyzed degree of body fat consumption. The analysis result of the learning model 131_1 may be calculated by compensating the pattern of the concentration signals SD1 to SDn with the physical quantity signals HD and CD. Through the nonlinear machine learning-based learning model 131_1 , a pattern of concentrations corresponding to body fat consumption and a pattern of concentrations corresponding to maintenance or increase of body fat may be distinguished. In addition, since moisture and impurities in the exhaled air cannot be completely trapped in the gas sensor array 120 , the physical quantity signals HD and CD by the physical quantity detection sensor 110 may be further considered for pattern recognition. That is, the learning model 131_1 is implemented to compensate for a hardware error of the gas sensor array 120 and to improve the analysis accuracy of body fat consumption by using various biomarkers.

4 is an exemplary block diagram of the gas sensor array of FIG. 1 ; The gas sensor array 120_2 of FIG. 4 will be understood as an exemplary embodiment of the gas sensor array 120 of FIG. 1 different from the embodiment of FIG. 2 . Referring to FIG. 4 , the gas sensor array 120_2 may include a plurality of gas sensors SE1 , SE2 , ?? and SEn , a resistance meter 121 , a converter 122 , and a gas discriminator 123 . can

Each of the plurality of gas sensors SE1 to SEn may detect a target substance (eg, acetone) in the exhalation of the object. The resistance measuring device 121 may measure electrical resistance that changes in each of the plurality of gas sensors SE1 to SEn according to a reaction with the target material. The converter 122 generates concentration signals SD1, SD2, ??, SDn of the plurality of gas sensors SE1 to SEn based on the measured electrical resistance of each of the gas sensors SE1 to SEn. can do. The configuration of the plurality of gas sensors SE1 to SEn, the resistance meter 121, and the converter 122 includes the plurality of gas sensors SE1 to SEn described in FIG. 2, the resistance meter 121, and the converter ( 122), a detailed description thereof will be omitted.

The gas discriminator 123 may receive the concentration signals SD1 to SDn and analyze the concentration of the target substance in the exhaled air based on the machine learning-based learning model 124 . Here, the learning model 124 is managed by the gas discriminator 123 of the gas sensor array 120_2 and may be a machine learning model separate from the learning model 131 described in FIG. 1 . The gas discriminator 123 may analyze the concentration signals SD1 to SDn using the learning model 124 to generate a reaction signal AO of the target material and a non-reaction signal AX of the target material. As an example, the reaction signal AO of the target material may represent a predicted value of the concentration of acetone in the expiration date, and the non-reaction signal AX of the target material may represent the predicted value of the concentration of gas other than acetone in the expiration. That is, the gas discriminator 123 may perform an intermediate analysis of the acetone concentration detected by the plurality of gas sensors SE1 to SEn before applying the physical quantity signals to determine body fat consumption.

The learning model 124 may be implemented in the gas discriminator 123, and may be trained to non-linearly analyze the concentration of the target material in the exhaled air from the concentration signals SD1 to SDn. The learning model 124 may be integrated and managed within the body fat analysis system 100 of FIG. 1 , but is not limited thereto, and the learning model 124 is implemented in a separate server or storage medium, and then determines the gas during analysis. It may be loaded into the group 123 . The learning model 124 may be built through an artificial neural network or deep learning machine learning. As an example, the learning model 124 is a convolutional neural network (CNN), a recurrent neural network (RNN), a deep neural network (DNN), a K-Nearest Neighbor (KNN), and decision making. It may be composed of at least one of a decision tree (DT), a multi-layer perceptron (MLP), a partial least square method (PLS), and a support vector machine.

FIG. 5 is an exemplary diagram for explaining the operation of the learning model of FIG. 4 . The learning model 124 of FIG. 5 will be understood as an exemplary embodiment of the learning model 124 of FIG. 4 . As described with reference to FIG. 4 , the learning model 124 may analyze the concentration of the target substance in the exhaled air based on the concentration signals SD1 to SDn.

The learning model 124 may be implemented as a machine learning-based neural network. The learning model 124 may be implemented as an input layer (IL), a hidden layer (HL), and an output layer (OL). It will be understood that these configurations are exemplary, and the learning model 124 may be implemented with a layer structure different from that of FIG. 5 according to the type or shape of the neural network. The input layer IL may receive the concentration signals SD1 to SDn. The received concentration signals SD1 to SDn may propagate to the hidden layer HL. According to learning, a weight parameter may be applied to each signal propagated to the hidden layer HL. The learning model 124 may calculate a relationship between the concentration signals SD1 to SDn based on the weight parameter. The output layer OL may output the analysis result of the concentration of the target substance in the air based on the analysis result propagated from the hidden layer HL.

The analysis result of the learning model 124 may include a response signal AO of the target material and a non-response signal AX of the target material. Through the non-linear machine learning-based learning model 124, a pattern of acetone concentrations in exhalation and a pattern of gas concentrations other than acetone in exhalation can be distinguished. In addition, moisture and impurities in the exhaled air, and the physical properties of each of the plurality of gas sensors SE1 to SEn may be considered for pattern recognition of the acetone concentration in the exhaled air.

FIG. 6 is an exemplary view for explaining the operation of the learning model of FIG. 1 to which the gas sensor array of FIG. 4 is applied. The learning model 131_2 of FIG. 6 will be understood as an exemplary embodiment of the learning model 131 of FIG. 1 when the gas sensor array 120 of FIG. 1 is the gas sensor array 120_2 of FIG. 4 .

Unlike FIG. 3 , the learning model 131_2 is the result of intermediate analysis by the learning model 124 of FIGS. 4 and 5 , instead of the concentration signals SD1 to SDn , the response signal AO of the target material and the target material. It is possible to receive the non-response signal AX of The learning model 131_2 analyzes the user's body fat consumption or diet progress based on the physical quantity signals HD and CD, the response signal AO of the target substance, and the non-response signal AX of the target substance. can

The learning model 131_2 may be implemented as a machine learning-based neural network. The learning model 131_2 may be implemented as an input layer IL, a hidden layer HL, and an output layer OL. It will be understood that these configurations are exemplary, and the learning model 131_2 may be implemented in a layer structure different from that of FIG. 6 according to the type or shape of the neural network. The input layer IL may receive the physical quantity signals HD and CD, the response signal AO of the target material, and the non-response signal AX of the target material. The received signals are propagated to the hidden layer HL, and a weight parameter according to learning may be applied to the signals. The output layer OL may output an analysis result of the degree of body fat consumption or the degree of diet progress of the learning model 131_2 . The analysis result may include consumption information DO indicating the extent to which the user's body fat is consumed and non-consumption information DX indicating the extent to which the user's body fat is not consumed. Furthermore, the learning model 131_2 may further calculate a time when body fat consumption starts based on the analyzed degree of body fat consumption.

7 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention. Referring to FIG. 1 , the body fat analysis system 200 may include a physical quantity detection sensor 210 , a gas sensor array 220 , an analyzer 230 , and a communication module 240 . The body fat analysis system 200 of FIG. 7 is a wearable body fat analysis device capable of detecting a user's physical factor and a target substance in exhalation, and analyzing the factors detected with various biomarkers to provide the user with a body fat consumption or diet progress status. can be implemented as

The physical quantity detection sensor 210 may detect a physical factor of an object. The physical quantity detection sensor 210 may detect at least one physical factor of the object. For example, the physical quantity detection sensor 210 includes a heart rate sensor 211 that detects the heartbeat of an object to generate a heartbeat signal HD, and a momentum sensor 112 that detects an exercise amount of the object and generates a momentum signal CD. may include The physical quantity detecting sensor 210 including the heart rate sensor 211 and the exercise amount sensor 212 corresponds to the physical quantity detecting sensor 110 including the heart rate sensor 111 and the exercise amount sensor 112 of FIG. 1 .

The gas sensor array 220 may detect a target substance in the exhalation of an object. The gas sensor array 220 may include a plurality of gas sensors SE, a resistance meter 221 , a converter 222 , and a gas discriminator 223 . The plurality of gas sensors SE, the resistance meter 221 , the converter 222 , and the gas discriminator 223 are the gas sensor array 120_2 of FIG. 4 is the plurality of gas sensors SE1 to SEn, resistance It corresponds to the measuring instrument 121 , the converter 122 , and the gas discriminator 123 . As shown in FIG. 4 , the gas sensor array 220 may analyze the concentration of the target material in the exhaled air based on the learning model implemented in the gas discriminator 223 . The gas sensor array 220 analyzes the concentrations detected by each of the plurality of gas sensors SE with the learning model 124 implemented in the gas discriminator 123 , and the response signal AO of the target material and the target material can generate an unresponsive signal (AX) of

The analyzer 230 receives the physical quantity signals HD and CD from the physical quantity detection sensor 110 , and receives a reaction signal AO of the target material and a non-response signal AX of the target material from the gas sensor array 220 . can receive The analyzer 230 may apply a linear regression equation to the physical quantity signals HD and CD and the concentration signals SD1 to SDn based on the linear analysis model 231 . That is, the analyzer 230 may analyze the user's body fat consumption or diet progress by using a quantified regression equation without using a machine learning-based learning model. The linear analysis model 231 may be used to classify whether or not body fat is consumed. Each component of the analyzer 230 may be implemented in hardware, firmware, software, or a combination thereof.

The communication module 240 may be configured to allow the body fat analysis system 200 to communicate with another external electronic device through a wired or wireless network. The communication module 240 corresponds to the communication module 140 of FIG. 1 .

8 is an exemplary block diagram of a body fat analysis system according to an embodiment of the present invention. Referring to FIG. 8 , the body fat analysis system 300 includes a plurality of gas sensors SE1, SE2, ??, and SEn, a resistance meter 310 , a converter 320 , an analyzer 330 , and a communication module 340 . ) may be included. The body fat analysis system 300 of FIG. 8 may be implemented as a wearable body fat analysis device capable of detecting a target substance in the user's exhalation and analyzing the concentrations of the detected target substance to provide the user with a body fat consumption or diet progress status. have.

Each of the plurality of gas sensors SE1 to SEn may detect a target substance (eg, acetone) in the exhalation of the object. The resistance meter 310 may measure electrical resistance that changes in each of the plurality of gas sensors SE1 to SEn according to a reaction with a target material. The converter 320 generates the concentration signals SD1, SD2, ??, SDn of the plurality of gas sensors SE1 to SEn based on the measured electrical resistance of each of the gas sensors SE1 to SEn. can do. The configuration of the plurality of gas sensors SE1 to SEn, the resistance meter 310, and the converter 320 includes the plurality of gas sensors SE1 to SEn described in FIG. 2, the resistance meter 121, and the converter ( 122), a detailed description thereof will be omitted.

The analyzer 330 may analyze the concentration signals SD1 to SDn based on the machine learning-based learning model 331 . As a result of the analysis of the learning model 331 , an analysis result indicating a degree of body fat consumption or a degree of diet progress of the user may be calculated. Each component of the analyzer 330 may be implemented in hardware, firmware, software, or a combination thereof.

The learning model 331 may be implemented in the analyzer 330 and learned to non-linearly analyze the user's body fat consumption or diet progress from the concentration signals SD1 to SDn. The learning model 331 may be integrated and managed within the body fat analysis system 300, but is not limited thereto, and the learning model 331 is implemented in a separate server or storage medium, and is then transferred to the analyzer 330 during analysis. can be loaded. The learning model 331 may be built through an artificial neural network or deep learning machine learning. As an example, the learning model 331 may include a convolutional neural network (CNN), a recurrent neural network (RNN), a deep neural network (DNN), a K-Nearest Neighbor (KNN), and decision making. It may be composed of at least one of a decision tree (DT), a multi-layer perceptron (MLP), a partial least square method (PLS), and a support vector machine.

The communication module 340 may be configured to allow the body fat analysis system 300 to communicate with another external electronic device through a wired or wireless network. The communication module 340 corresponds to the communication module 140 of FIG. 1 .

FIG. 9 is an exemplary diagram for explaining the operation of the learning model of FIG. 8 . The learning model 331 of FIG. 9 will be understood as an exemplary embodiment of the learning model 331 of FIG. 8 . As described with reference to FIG. 8 , the learning model 331 may analyze the user's body fat consumption or diet progress based on the concentration signals SD1 to SDn.

The learning model 331 may be implemented as a machine learning-based neural network. The learning model 331 may be implemented as an input layer (IL), a hidden layer (HL), and an output layer (OL). It will be understood that these configurations are exemplary, and the learning model 331 may be implemented with a layer structure different from that of FIG. 9 according to the type or shape of the neural network. The input layer IL may receive the concentration signals SD1 to SDn. The received concentration signals SD1 to SDn may propagate to the hidden layer HL. According to learning, a weight parameter may be applied to each signal propagated to the hidden layer HL. The learning model 331 may calculate a relationship between each signal based on the weight parameter. The output layer OL may output the analysis result of the degree of body fat consumption or the degree of diet progress based on the analysis result propagated from the hidden layer HL.

The analysis result of the learning model 331 shows consumption information DO indicating the extent to which the user's body fat is consumed (the extent to which the diet has progressed) and the non-consumption information DO indicating the extent to which the user's body fat is not consumed (the extent to which the diet is not progressed). Information DX may be included. Furthermore, the learning model 331 may further calculate a time when body fat consumption starts based on the analyzed degree of body fat consumption. Through the nonlinear machine learning-based learning model 331 , a pattern of concentrations corresponding to body fat consumption and a pattern of concentrations corresponding to maintenance or increase of body fat may be distinguished.

10 is an exemplary block diagram of an electronic system to which the body fat analysis system described in FIGS. 1 to 9 is applied. Referring to FIG. 10 , the electronic system 400 includes a gas sampler 410 , a gas sensor array 420 , a signal processor 430 , a fluid controller 440 , an environment controller 450 , an environment sensor 460 , and communication. module 470 , and a controller 480 . The electronic system 400 of FIG. 10 is exemplary, and the electronic system 400 does not include at least some of the components of FIG. 10 , or includes the physical quantity detection sensors 110 and 210 described in FIGS. 1 to 9 . may include more.

The gas sampler 410 may be configured to trap at least one material different from the target material and transfer the remaining materials to the gas sensor array 420 . For example, the gas sampler 410 may remove or filter substances other than acetone, such as moisture. Accordingly, the accuracy of the target material detected by the gas sensor array 420 may be improved.

The gas sensor array 420 may detect a target substance in the object's exhalation. The gas sensor array 420 may include a plurality of gas sensors SE1, SE2, ??, and SEn for detecting a target material. The gas sensor array 420 may be accommodated in the reaction chamber. The reaction chamber may be formed so that other gases other than the gas introduced through the gas sampler 410 do not flow into the gas sensor array 420 . The gas sensor array 420 includes a concentration calculator that measures changes in electrical resistance by a target material of each of the plurality of gas sensors SE1, SE2, ??, and SEn, and generates concentration signals based on the changes. can do. This concentration calculator will be understood as a concept including the above-described resistance meter (121, 221, 310) and converters (122, 222, 320).

The signal processor 430 may analyze the concentration signals of the target material generated from the gas sensor array 420 to generate an analysis result indicating the degree of body fat consumption or diet progress. The signal processor 430 may include the analyzers 130 , 230 , and 330 described above. 1 or 8 , the signal processor 430 may non-linearly analyze the degree of body fat consumption or diet progress by using a machine learning-based learning model. According to the embodiment of FIG. 7 , the signal processor 430 may linearly analyze the degree of body fat consumption or diet progress using a linear regression equation.

The signal processor 430 may operate under the control of the controller 480 and may include a working memory for signal processing using a learning model or the like. As an example, the learning model may be implemented in firmware or software. For example, the firmware may be loaded into the working memory of the signal processor 430 when executed by the controller 480 . The controller 480 may execute the loaded firmware. Under the control of the controller 480 , the signal processor 430 may analyze the user's body fat consumption or diet progress using the learning model. However, the present invention is not limited thereto, and the signal processor 430 may include a neural network circuit in which the learning model is implemented in hardware.

The fluid controller 440 may control the flow of gas provided to the gas sensor array 420 . In one example, fluid controller 440 may include a valve, pump, filter, or orifice for controlling the flow of gas within the chamber.

The environment controller 450 may control the environment of the gas provided to the gas sensor array 420 . For example, the environmental controller 450 may include a heater (or cooler) for temperature control, a dehumidifier for humidity control, or an air pressure control device for atmospheric pressure control.

The environment sensor 460 may sense the environment of the gas provided to the gas sensor array 420 . The environment controller 450 may control the environment of the aircraft under the control of the controller 480 based on the environment information sensed by the environment sensor 460 . For example, the environmental sensor 460 may include a thermometer for temperature measurement, a hygrometer for humidity measurement, or a pressure sensor for atmospheric pressure measurement.

The communication module 470 may be configured to allow the electronic system 400 to communicate with another external electronic device through a wired or wireless network. The communication module 470 corresponds to the above-described communication modules 140 , 240 , 340 .

The controller 480 may function as a central processing unit of the electronic system 400 . The controller 480 includes a gas trap of the gas sampler 410 , generation of concentration signals of the gas sensor array 420 , analysis of body fat consumption by the signal processor 430 , gas flow control of the fluid controller 440 , and an environment controller It is possible to control the environment control of the aircraft at 450 , the environment detection of the gas by the environment sensor 460 , and signal transmission/reception of the communication module 470 .

The contents described above are specific examples for carrying out the present invention. The present invention will include not only the above-described embodiments, but also simple design changes or easily changeable embodiments. In addition, the present invention will also include techniques that can be easily modified and implemented in the future using the above-described embodiments.

100, 200, 300: body fat analysis system
110, 210: physical quantity detection sensor
120, 120_1, 120_2, 220, 420: gas sensor array
SE, SE1-SEn: gas sensor
121, 221, 310: resistance meter
122, 222, 320: converter
123, 223: gas discriminator
130, 230, 330: analyzer
140, 240, 340, 470: communication module

Claims (20)

a physical quantity detection sensor that detects a physical factor and generates a physical quantity signal;
A gas sensor array comprising a plurality of gas sensors for detecting a target material in air, and generating concentration signals of the target material respectively corresponding to the detection signals based on detection signals generated from the plurality of gas sensors ; and
and an analyzer for generating body fat consumption information by analyzing each of the physical quantity signal and the concentration signal through a machine learning model. According to claim 1,
The physical quantity detection sensor,
a heartbeat sensor detecting a heartbeat of an object to generate a first physical quantity signal; and
and a momentum sensor that detects the momentum of the object and generates a second physical quantity signal,
The analyzer analyzes each of the first and second physical quantity signals and the concentration signals through the machine learning model to generate the body fat consumption information. According to claim 1,
The physical quantity detection sensor,
At least one physical sensor for detecting the physical factor,
The number of the plurality of gas sensors is greater than the number of the at least one physical sensor. According to claim 1,
The target material is acetone. According to claim 1,
Each of the plurality of gas sensors, a body fat analysis system having different reactivity to the target material. According to claim 1,
The gas sensor array,
a resistance meter for measuring an electrical resistance of each of the plurality of gas sensors by the target material based on the detection signals; and
and a converter for calculating concentrations of the target material detected from the plurality of gas sensors based on the measured electrical resistance and generating the concentration signals based on the calculated concentrations. According to claim 1,
and a gas sampler that traps at least one material different from the target material and passes the target material through the gas sensor array so that the target material is transmitted to the gas sensor array. According to claim 1,
a fluid controller controlling the flow of air delivered to the gas sensor array;
an environment controller for controlling an environment of air delivered to the gas sensor array; and
The body fat analysis system further comprising an environmental sensor for sensing the environment. a physical quantity detection sensor that detects a physical factor and generates a physical quantity signal;
A target material response signal comprising a plurality of gas sensors for detecting a target material in the air, calculating concentrations of the target material detected from the plurality of gas sensors, and analyzing each of the concentrations through a machine learning model and a gas sensor array for generating a target material non-responsive signal; and
and an analyzer configured to generate body fat consumption information based on the physical quantity signal, the target substance response signal, and the target substance non-response signal. 10. The method of claim 9,
The physical quantity detection sensor,
At least one physical sensor for detecting the physical factor,
The number of the plurality of gas sensors is greater than the number of the at least one physical sensor. 10. The method of claim 9,
The target material is acetone,
A first gas sensor and a second gas sensor among the plurality of gas sensors have different reactivity to acetone. 10. The method of claim 9,
The electrical resistance of each of the plurality of gas sensors is changed by the target material,
The gas sensor array is configured to calculate concentrations of the target material based on a change in the electrical resistance. 10. The method of claim 9,
The analyzer is configured to generate the body fat consumption information by applying a weight parameter learned to the physical quantity signal, the target substance response signal, and the target substance non-response signal. 10. The method of claim 9,
The analyzer is configured to generate the body fat consumption information by applying a linear regression equation to the physical quantity signal, the target substance response signal, and the target substance non-response signal. A wearable body fat analyzer comprising:
a plurality of gas sensors for detecting a target substance contained in the user's exhalation;
a concentration calculator configured to calculate a concentration of a target material detected from each of the plurality of gas sensors based on the reactivity of each of the plurality of gas sensors to the target material; and
and an analyzer configured to determine body fat consumption by analyzing the concentrations of the target substances respectively detected from the plurality of gas sensors through a machine learning model. 16. The method of claim 15,
The electrical resistance of each of the plurality of gas sensors is changed by the target material,
The concentration calculator is configured to calculate the concentrations of the target material respectively corresponding to the plurality of gas sensors, based on a change in the electrical resistance. 16. The method of claim 15,
The analyzer analyzes the concentrations of the target substance, and determines a start time of body fat consumption. 16. The method of claim 15,
A first gas sensor and a second gas sensor among the plurality of gas sensors have different reactivity with respect to the target material. 16. The method of claim 15,
Further comprising a physical quantity detection sensor that detects a physical factor to generate a physical quantity signal,
The analyzer further inputs the physical quantity signal to the machine learning model to determine the body fat consumption. 20. The method of claim 19,
The physical quantity detection sensor,
At least one physical sensor for detecting the physical factor,
The number of the plurality of gas sensors is greater than the number of the at least one physical sensor.

Images