JP6425393B2 - Prediction system, prediction method, and prediction program - Google Patents

Description

  One aspect of the present invention relates to a prediction system, a prediction method, and a prediction program that use machine learning to estimate the living behavior of the subject.

  Eating and drinking is one of the most important things in people's daily life, and it is also important in health management. For example, in the nursing or care setting, managing the diet of the patient or the cared person is essential. Conventionally, techniques for supporting management of such meals are known. For example, according to Patent Document 1 below, there is a life management terminal for detecting that the user is eating based on information measured by the user, and a predetermined after detection when the start or end of the eating is detected. A lifestyle management system is described which comprises means for inquiring the user of the meal content after time and means for obtaining an answer from the user to the inquiry.

Unexamined-Japanese-Patent No. 2003-173375

  In managing food, it is necessary not only to know whether a drink or food has been taken, but also to understand its intake, but it is not easy to observe its intake. Even knowing the intake of one person's single meal is difficult, let's say, the intake of each person's meal at the hospital or care facility that cares for multiple (several or many) patients or carers It is very difficult to grasp the quantity. Therefore, it is desirable to easily grasp the intake of the subject.

  The prediction system according to one aspect of the present invention generates an first conversion data from the biological signal and an acquisition unit that acquires a biological signal generated by the swallowing motion of the subject, and uses the first conversion data as an explanatory variable. The second conversion data from the biological signal when it is determined that the subject has taken the ingested item by performing the first machine learning, and if it is determined that the subject has taken the subject By performing second machine learning using the second transformation data as an explanatory variable to estimate an intake amount of an intake of the subject, and an output for outputting the estimated intake amount. And a unit.

  The prediction method according to an aspect of the present invention is a prediction method executed by a prediction system including a processor, which includes an acquisition step of acquiring a biological signal generated by the swallowing motion of the subject, and a first conversion from the biological signal. A determination step of determining whether the subject has taken an intake by performing data generation and first machine learning using the first conversion data as an explanatory variable; If it is determined that the second transformation data has been determined, the second transformation data is generated from the biological signal, and the second machine learning is performed using the second transformation data as an explanatory variable, whereby the intake amount of the intake of the subject is obtained. The estimation step includes estimation, and an output step of outputting the estimated intake.

  The prediction program according to one aspect of the present invention generates an first conversion data from the biological signal and an acquisition unit that acquires a biological signal generated by the swallowing motion of the subject, and uses the first conversion data as an explanatory variable. The second conversion data from the biological signal when it is determined that the subject has taken the ingested item by performing the first machine learning, and if it is determined that the subject has taken the subject By performing second machine learning using the second transformation data as an explanatory variable to estimate an intake amount of an intake of the subject, and an output for outputting the estimated intake amount. Make the computer function as a part.

  In such an aspect, first, the presence or absence of intake is determined by performing the first machine learning based on the biological signal generated by the swallowing motion of the subject. Then, if it is determined that the subject has taken it, the intake is estimated by the second machine learning based on the biological signal. As described above, by performing machine learning on the biological signal generated by the swallowing movement, it is possible to easily grasp the intake of the subject.

  According to one aspect of the present invention, the intake of the subject can be easily grasped.

It is a flow chart which shows the concept of operation of a prediction system concerning an embodiment. It is a figure which shows the hardware constitutions of the computer which comprises the prediction system which concerns on embodiment. It is a block diagram showing functional composition of a prediction system concerning an embodiment. It is a figure which shows the example of the power spectral density obtained from biosignal data. It is a figure showing division of living body signal data. It is a figure which shows the example of the biosignal data which show one swallowing area. It is a figure which shows the example of the amplitude histogram obtained from biosignal data. It is a flow chart which shows operation of a prediction system concerning an embodiment. It is a figure showing composition of a prediction program concerning an embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. In the description of the drawings, the same or equivalent elements will be denoted by the same reference symbols, without redundant description.

  First, the functions and configuration of the prediction system 10 according to the embodiment will be described using FIGS. The prediction system 10 is a computer system that uses machine learning to perform estimation regarding the oral intake of the subject.

  Machine learning is a process of generating a pattern function by learning training data which is a set of known values, and predicting an unknown value using the pattern function. The type of machine learning used by the prediction system 10 is not limited. Examples of machine learning include artificial neural networks (ANN), support vector machines (SVM), decision tree learning, correlation rule learning, Bayesian networks, etc., but other algorithms may be used.

  The "subject" in the present specification is a person who is a target to observe intake of an intake. Any person may be observed regardless of gender or age. The term "intake" as used herein refers to any drink or food, and the type is not limited in any way. As used herein, "intake" is the act of taking the ingestible through the mouth, decimating the ingestible if necessary, and flushing the ingestible into the stomach (ie, oral ingestion).

  An overview of intake estimation using the prediction system 10 is shown in FIG. First, data of biological signals generated by the swallowing movement of the subject is collected (step S1). Subsequently, the presence or absence of intake is determined by first machine learning based on the biological signal (step S2). Subsequently, the biological signal is divided (step S3), and a single bite intake based on each divided signal is estimated using the second machine learning (step S4). Then, a desired total intake amount is calculated from the individual bite intake amounts (step S5).

  Here, the swallowing exercise is an exercise that occurs in the body of the subject in the process from the intake of the intake into the mouth to the stomach. For example, chewing and swallowing are a type of swallowing exercise. The “biosignal generated in the swallowing movement” in the present specification is a signal generated in the body of the subject during the swallowing movement. Specific examples of the signal include sound (depression sound), vibration, potential change during muscle activity (myoelectricity), and the like, but the type of signal is not limited to these.

  The biological signal is detected by a sensor attached to the subject, and stored in the database as biological signal data (collection of biological signal data). The biological signal sensor is a device that detects a biological signal. As described above, since the type of biological signal is not limited, the type of sensor that captures the signal is also not limited. The sensor may be attached to the subject (a so-called wearable sensor) or may be provided anywhere in the room. As an example, it is conceivable to use a throat microphone as a sensor if it captures swallowing sound, but this microphone is an aspect of a wearable sensor. When wearable sensors are used, it is possible to collect biosignals without restricting the behavior of the subject (for example, without restricting the place and time of eating and drinking).

  "Bite intake" is the amount of intake taken into the stomach from the mouth in a single swallow, in other words, the amount of intake taken in one swallow. The type of amount is not limited, and it may be volume, mass, weight, heat, or any other amount. The total intake amount is a total value of a single intake amount in any period, and may be, for example, an intake amount in one meal or an intake amount in one day.

  The desired total intake can be information used for various purposes. For example, the intake status of the subject can be grasped from the total intake, and the information can be used for health management of the subject, planning of future meals, and the like.

  When the prediction system 10 includes one or more computers and a plurality of computers, each functional element of the prediction system 10 described later is realized by distributed processing. The type of individual computer is not limited. For example, a stationary or portable personal computer (PC) may be used, or a workstation may be used, or a mobile terminal such as a high-performance mobile phone (smart phone), a mobile phone, or a personal digital assistant (PDA). May be used. Alternatively, various types of computers may be combined to construct the prediction system 10. When using a plurality of computers, these computers are connected via a communication network such as the Internet or an intranet.

  The general hardware configuration of the individual computers 100 in the prediction system 10 is shown in FIG. The computer 100 includes a CPU (processor) 101 that executes an operating system, application programs, and the like, a main storage unit 102 configured by a ROM and a RAM, an auxiliary storage unit 103 configured by a hard disk and a flash memory, and the like. The communication control unit 104 includes a card or a wireless communication module, an input device 105 such as a keyboard or a mouse, and an output device 106 such as a display or a printer. Of course, the hardware modules mounted differ depending on the type of computer 100. For example, while stationary PCs and workstations often include a keyboard, a mouse, and a monitor as input devices and output devices, in a smartphone, a touch panel often functions as the input devices and output devices.

  Each functional element of the prediction system 10 to be described later reads predetermined software on the CPU 101 or the main storage unit 102, and operates the communication control unit 104, the input device 105, the output device 106, etc. under the control of the CPU 101, This is realized by reading and writing data in the main storage unit 102 or the auxiliary storage unit 103. Data and databases necessary for processing are stored in the main storage unit 102 or the auxiliary storage unit 103.

  The prediction system 10 accesses a database 20 to obtain biosignal data, which is an apparatus or functional element that stores biosignal data. The implementation method of the database 20 is not limited. For example, database 20 may be in prediction system 10 or may be in a separate system from prediction system 10. Also, the database 20 may be a relational database or a CSV file. The method of accumulating the biosignal data in the database 20 is not limited. For example, the sensor S may communicate directly with the database 20 and store the biosignal data in the database 20, or another computer collects the biosignal data from the sensor S and stores the biosignal data in the database 20 May be

  As shown in FIG. 3, the prediction system 10 includes an acquisition unit 11, a determination unit 12, an estimation unit 13, and an output unit 14 as functional components. Below, a biological signal is swallowing sound and the process in case the prediction system 10 estimates the water intake of a to-be-observed person from the swallowing sound is shown as an example. Here, the amount of water intake is the amount of intake of a drink (for example, water) by the subject.

  The acquisition unit 11 is a functional element that acquires a biological signal generated by the swallowing motion of the subject. In general, biosignals are collected continuously over some time span (eg, 1 second, 1 minute, 1 hour, etc.) and can be represented as waves (see FIG. 1). The biological signal is detected by a sensor (for example, a throat microphone) S mounted on the subject U, and is stored in the database 20 as biological signal data. The acquisition unit 11 reads the biological signal data from the database 20 and outputs the data to the determination unit 12. The time width of the biological signal data acquired by the acquisition unit 11 for a certain subject is not limited at all. For example, the time width may be 30 seconds, 1 minute, 10 minutes, 30 minutes, 1 hour, or 2 hours.

  The determination unit 12 is a functional element that estimates whether the subject has ingested an intake (drink). The determination unit 12 generates the first conversion data by performing the first conversion on the biological signal data. Then, the determination unit 12 executes the first machine learning on the first conversion data to estimate whether the subject has ingested the intake (drink). The data conversion method and machine learning used to determine the presence or absence of intake are not limited at all. In this embodiment, the determination unit 12 uses fast Fourier transform (FFT) as a data conversion method, and uses support vector classification (SVC) in which SVM corresponds to classification as machine learning.

  The determination unit 12 performs fast Fourier transform (FFT) on the biological signal data to obtain the power spectral density (PSD) of the swallowing sound as first converted data. Subsequently, the determination unit 12 executes SVC using the power spectral density as an explanatory variable to determine whether the input biological signal data indicates an action of "drink".

  The power spectral density obtained from swallowing sound data is characteristic of various behaviors using the throat. FIG. 4 shows power spectral densities for six types of “drink”, “eat”, “cough”, “breath”, “speak” and “spit”. The vertical and horizontal axes of each graph in FIG. 4 are intensity and frequency, respectively. As can be seen from these graphs, it can be seen that the power spectrum density of each action has features different from those of the other actions.

  Not all the biosignal data collected using the sensor S is generated by the "drink" action, but there is a possibility that there is biosignal data generated by another action necessary for a person to live. Therefore, as processing prior to estimating the water intake, the determination unit 12 analyzes the power spectral density with SVC, and the input biological signal data indicates an act of "drinking" or "other than drinking" Determine whether to indicate an act. Then, when it is determined that the biosignal data indicates the act of "drinking" (that is, when it is determined that the subject drinks the drink), the estimation unit 13 outputs the biosignal data to the estimation unit 13. Do. On the other hand, when it is determined that the biosignal data indicates an action other than "drink" (that is, when it is determined that the subject is not drinking a drink), the estimation unit 13 estimates the biosignal data. In this case, the processing of the estimation unit 13 and the output unit 14 described later is not performed.

  A predetermined number (for example, 255) of explanatory variables can be obtained as frequency characteristics of power spectral density obtained by FFT. The determination unit 12 may use all of the explanatory variables or may use only some of the explanatory variables necessary for estimation. Since all explanatory variables do not necessarily contribute to the improvement of the accuracy of judgment, it is possible to improve the accuracy of judgment by executing SVC after selecting only the necessary explanatory variables, and it is also expected to reduce the calculation time. it can. There are two types of methods for selecting the explanatory variable, an increase method and a decrease method.

  The increase method is a method in which processing is started from a model with zero explanatory variables, and one explanatory variable with the highest correlation value X is added to the target variable one by one. In this method, the initial value of the explanatory variable group A used for machine learning is null, and the initial value of the explanatory variable group B not used for machine learning is N (for example, N = 255) of explanatory variables. It is.

  First, the determination unit 12 obtains the correlation value X with respect to the objective variable for each explanatory variable in the explanatory variable group B, and transfers the explanatory variable with the highest correlation value X from the explanatory variable group B to the explanatory variable group A. Subsequently, the determination unit 12 performs regression analysis on the explanatory variable group A to confirm estimated coefficients of the respective explanatory variables. Then, when the absolute value Y of the estimation coefficient is 1 or more for all explanatory variables, the determination unit 12 returns to the process of transferring the explanatory variable having the highest correlation value X to the explanatory variable group A, and the subsequent process. repeat. On the other hand, when an explanatory variable whose absolute value Y of the estimation coefficient falls below 1 appears, the determination unit 12 executes machine learning using the explanatory variable group A after the explanatory variable added last is removed.

  On the other hand, the reduction method is a method in which processing is started from a model adopting all explanatory variables, and one explanatory variable with the lowest correlation value X with respect to the target variable is removed one by one. In this method, the determination unit 12 determines that N initial values of the explanatory variable group A used for machine learning are N (for example, N = 255) and the initial values of the explanatory variable group B not used for machine learning are empty. (Null)

  In the present embodiment, first, the number of explanatory variables in the explanatory variable group A is reduced from N to Na (for example, reduced from 255 to 15). For this purpose, the determination unit 12 obtains the correlation value X with respect to the objective variable for each explanatory variable of the explanatory variable group A, and transfers the explanatory variable with the lowest correlation value X from the explanatory variable group A to the explanatory variable group B. The determination unit 12 repeats this movement process until the number of explanatory variables in the explanatory variable group A becomes Na.

  Subsequently, the determination unit 12 performs regression analysis with the explanatory variable group A to confirm estimated coefficients of the respective explanatory variables. Then, when the absolute value Y of the estimation coefficient is 1 or more for all explanatory variables, the determination unit 12 executes machine learning using the explanatory variable group A. On the other hand, when an explanatory variable in which the absolute value Y of the estimation coefficient falls below 1 appears, the determination unit 12 transfers the explanatory variable with the lowest correlation value X to the explanatory variable group B, and repeats the processing after the regression analysis.

  The estimation unit 13 is a functional element that estimates the intake of the intake of the subject. The estimation unit 13 generates second conversion data by performing the second conversion on the biological signal data. Then, the estimation unit 13 estimates the intake amount by executing the second machine learning on the second conversion data. The data conversion method and machine learning used to estimate the intake amount are not limited at all. In the present embodiment, the estimation unit 13 uses the amplitude histogram as a data conversion method, and uses SVC as machine learning. The second conversion data is data indicating characteristics (ie, time characteristics) of values that change with the passage of time, and the amplitude histogram is a type of data indicating the time characteristics. Since data indicating characteristics of the amount of change of adjacent waves in the biological signal data is also one type of data indicating time characteristics, data indicating the characteristics of the amount of change may be used instead of the amplitude histogram. Another example of the machine learning in the estimation unit 13 is support vector regression (SVR) in which SVM corresponds to regression.

  First, the estimation unit 13 specifies one or more swallowing intervals by dividing a time interval indicated by the biological signal data into a swallowing interval and a noise interval. Here, the swallowing section is a section estimated to indicate one swallowing, and the noise section is a section estimated that no swallowing has occurred. For this process, the estimation unit 13 holds in advance the maximum value V of the sound pressure level of the sensor S in the noise section and the minimum value T of the duration of the noise section as a threshold. These threshold values V and T are set in advance by the manager of the prediction system 10. The estimation unit 13 determines that a section in which a waveform less than the threshold V continuously continues for the threshold T or more in the biological signal data is a noise section, and determines one or more sections remaining after excluding the noise section as a swallowing section Do. For example, as illustrated in FIG. 5, the estimation unit 13 divides the biosignal data into one or more swallowing intervals td and one or more noise intervals tn.

Subsequently, the estimation unit 13 estimates bite intake in each swallowing interval. When SVC is used as in the present embodiment, the estimation unit 13 prepares _n classes C _{1 to} C _n in advance for a single intake amount, and the single intake amount in each swallowing interval is C ₁ to C. _It is estimated which _n corresponds to. For example, four types of 5 ml, 10 ml, 15 ml and 20 ml may be prepared as classification of bite intake. Alternatively, the estimation unit 13 may estimate the bite intake by combining SVC and SVR. Specifically, the estimation unit 13 may first narrow down the estimation range with SVC and then estimate the bite intake in more detail using SVR. For example, when the bite intake estimated by SVC is 15 ml, the estimation unit 13 estimates the bite intake by SVR in the range of 15 ml ± 5 ml. The estimation unit 13 executes the following processing for each swallowing section.

  As illustrated in FIG. 6, the estimation unit 13 ignores a portion estimated to be noise (for example, a portion whose sound pressure level is less than the above-mentioned threshold V) vn in the sound pressure level range in the swallowing interval The other part (for example, the part where the sound pressure level is the above-mentioned threshold V or more) vd is equally divided into a plurality. The threshold for dividing the noise part and the other part in the swallowing section may be different from the threshold for dividing the biological signal data into the swallowing section and the noise section. The number of sound pressure level break positions is arbitrary, and may be 100 or 200, for example. In the example of FIG. 6, 200 division positions (c0 to c199) are set in the portion vd corresponding to the positive sound pressure level. Subsequently, the estimation unit 13 generates an amplitude histogram by counting how many times the wave of the sound pressure level has passed through each segment position (c0 to c199 in FIG. 6) (that is, counting the number of passes).

  The amplitude spectrum obtained from the swallowing sound data is characteristic of bite intake. FIG. 7 shows amplitude spectra for four types of 5 ml, 10 ml, 15 ml and 20 ml. The vertical axis and the horizontal axis of the graph of FIG. It can be seen that the greater the bite intake, the greater the number of passes at a particular location of the sound pressure level (see, for example, location R in FIG. 7). Therefore, it is possible to estimate bite intake from this amplitude histogram. The estimation unit 13 estimates the bite intake in the swallowing section to be processed by executing SVC with the amplitude histogram as an explanatory variable.

  The estimation unit 13 does not have to use the number of times of passage of all the break positions (c0 to c199 in FIGS. 6 and 7) when estimating the bite intake amount, and uses only some break positions necessary for the estimation process. May be In the example of FIG. 7, the characteristics of single-mouthed intake appear prominently over the range of dividing positions c0 to c40, and the number of times of passing becomes almost 0 regardless of single-mouthed intake in the range exceeding c40. I can not distinguish. Therefore, the estimation unit 13 may perform the SVC estimation using only the number of passes of a partial range (for example, c0 to c40) contributing to the classification. Since the amount of calculation is reduced by limiting the range of the amplitude histogram used for estimation in this manner, estimation processing can be performed faster.

  When the bite intake amount is estimated for all the swallowing sections, the estimation unit 13 outputs the estimation result to the output unit 14.

  The output unit 14 is a functional element that outputs the estimated intake as the estimation result 21. The output destination of the estimation result 21 is not limited at all. For example, the output unit 14 may display the estimation result 21 on a monitor or print it on a printer, may write the estimation result 21 in a text file, or store the estimation result 21 in a memory, a database, or the like. It may be stored in the device. Alternatively, the output unit 14 may output the estimation result 21 to another computer system (for example, a system for machine learning) via a communication network.

  The content and format of the estimation result 21 are also not limited. For example, the output unit 14 may output the single intake amount of each swallowing section which has been input, or the total value of the intake amount (for example, the total value of the single oral intake per one meal per day or per day) ) May be output. For example, when five single bit intake amounts “5 ml”, “15 ml”, “10 ml”, “5 ml” and “10 ml” are input, the output unit 14 outputs the total intake amount “45 ml” as the estimation result 21 Good. Alternatively, the output unit 14 may output both a single-mouth intake of each swallowing section and the total value of the intake.

  Next, the operation of the prediction system 10 will be described using FIG. 8 and the prediction method according to the present embodiment will be described.

  First, the acquisition unit 11 reads out biomedical signal data indicating swallowing noise from the database 20 (step S11, acquisition step). Subsequently, the determination unit 12 performs a fast Fourier transform (FFT) on the biological signal data to generate a power spectral density (PSD) of the swallowing sound (step S12). Subsequently, the determination unit 12 performs SVC on the power spectral density to determine the presence or absence of the intake ("drink" action) by the subject (step S13, determination step).

  If it is determined that the person to be observed has performed the action of “drinking” (step S14; YES), the estimation unit 13 divides the biological signal data to specify one or more swallowing sections (step S15). Then, for each swallowing section, the estimation unit 13 generates an amplitude histogram from the data of the swallowing section (step S16), and estimates the bite intake amount by executing SVC on the amplitude histogram (step S17, estimation) Step). The estimation unit 13 executes generation of an amplitude histogram and estimation of a single intake amount for all the swallowing intervals (see step S18). Finally, the output unit 14 outputs the estimation result of the intake (step S19, output step).

  On the other hand, when it is determined that the subject did not perform the action of “drinking” (step S14; NO), the prediction system 10 ends the processing on the acquired biological signal data without performing the estimation of the intake amount. Do.

  Next, a prediction program P1 for realizing the prediction system 10 will be described with reference to FIG.

  The prediction program P1 includes a main module P10, an acquisition module P11, a determination module P12, an estimation module P13, and an output module P14.

  The main module P10 is a part that centrally controls the prediction function by machine learning. The functions realized by executing the acquisition module P11, the determination module P12, the estimation module P13, and the output module P14 are the same as the functions of the acquisition unit 11, the determination unit 12, the estimation unit 13, and the output unit 14 described above, respectively. It is.

  The prediction program P1 may be provided, for example, after being fixedly recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. In addition, the prediction program P1 may be provided via a communication network as a data signal superimposed on a carrier wave.

  As described above, the prediction system according to one aspect of the present invention generates the first conversion data from the biological signal and an acquisition unit that acquires the biological signal generated by the swallowing motion of the subject, and performs the first conversion. A determination unit that determines whether the subject has ingested an intake by performing first machine learning using data as an explanatory variable, and a living body when it is determined that the subject has ingested An estimation unit configured to generate second transformation data from the signal and execute second machine learning using the second transformation data as an explanatory variable to estimate an intake amount of an intake of the subject; And an output unit that outputs an intake amount.

  The prediction method according to an aspect of the present invention is a prediction method executed by a prediction system including a processor, which includes an acquisition step of acquiring a biological signal generated by the swallowing motion of the subject, and a first conversion from the biological signal. A determination step of determining whether the subject has taken an intake by performing data generation and first machine learning using the first conversion data as an explanatory variable; If it is determined that the second transformation data has been determined, the second transformation data is generated from the biological signal, and the second machine learning is performed using the second transformation data as an explanatory variable, whereby the intake amount of the intake of the subject is obtained. The estimation step includes estimation, and an output step of outputting the estimated intake.

  The prediction program according to one aspect of the present invention generates an first conversion data from the biological signal and an acquisition unit that acquires a biological signal generated by the swallowing motion of the subject, and uses the first conversion data as an explanatory variable. The second conversion data from the biological signal when it is determined that the subject has taken the ingested item by performing the first machine learning, and if it is determined that the subject has taken the subject By performing second machine learning using the second transformation data as an explanatory variable to estimate an intake amount of an intake of the subject, and an output for outputting the estimated intake amount. Make the computer function as a part.

  In such an aspect, first, the presence or absence of intake is determined by performing the first machine learning based on the biological signal generated by the swallowing motion of the subject. Then, if it is determined that the subject has taken it, the intake is estimated by the second machine learning based on the biological signal. As described above, by performing machine learning on the biological signal generated by the swallowing movement, it is possible to easily grasp the intake of the subject.

  For example, the prediction system 10 can be used to keep track of each person's food intake at a hospital or care facility that cares for multiple (some or many) patients or carers. As health management including diet management needs to be performed continuously, manual operation not only burdens nurses and carers, but it may lead to omissions or errors in records. According to the present embodiment, the presence or absence and the intake amount are automatically output from the biological signal generated by the swallowing exercise, so that the burden of nursing or nursing care can be reduced or the intake state can be reliably recorded. It will be possible.

  In the prediction system according to the other aspect, the estimation unit estimates one or more swallowing segments by dividing the biological signal, estimates one intake amount of the ingested item for each swallowing interval, and the output unit includes a single intake amount, And at least one of the aggregate value of the bite intake may be output. By dividing the biosignal in this way and estimating the intake per bit, the intake can be estimated accurately.

  In a prediction system according to another aspect, the first transformation data may be power spectral density, and the first machine learning may be support vector classification.

  In a prediction system according to another aspect, the second transformation data may be an amplitude histogram, and the second machine learning may be support vector classification or a combination of support vector classification and support vector regression.

  In the prediction system according to another aspect, the biological signal may be swallowing sound. Since the swallowing sound can be easily detected by a microphone such as a throat microphone, it is possible to capture the swallowing movement while suppressing the discomfort of wearing given to the subject.

  The present invention has been described above in detail based on the embodiments. However, the present invention is not limited to the above embodiment. The present invention can be variously modified without departing from the scope of the invention.

  As described above, the specific techniques of the first and second machine learning are not limited. Also, the first machine learning and the second machine learning may be the same or different.

  In the above embodiment, the estimation unit estimates one or more swallowing segments by dividing the biological signal, and estimates a single bite intake amount for each swallowing segment, but dividing the biological signal into individual swallowing segments is not essential. . The prediction system may estimate a desired intake without using such division processing when it is determined that the subject has taken it.

  Reference Signs List 10 prediction system 11 acquisition unit 12 determination unit 13 estimation unit 14 output unit 20 database P1 prediction program P10 main module P11 acquisition module P12 determination module P13 ... estimation module, P14 ... output module.

Claims (8)

An acquisition unit for acquiring a biological signal generated by the swallowing motion of the subject;
The power spectral density is generated from the biological signal, the power spectral density is used as an explanatory variable , and the generated power spectral density is analyzed by first machine learning corresponding to classification of behavior using the throat. A determination unit that determines whether the subject has ingested an intake;
Wherein when it is determined that the viewer has ingested, time time characteristics data indicating the characteristic form the biological signal or al production, second corresponding to the time characteristic data as explanatory variables and intake classification An estimation unit configured to estimate the intake of the intake of the subject by performing machine learning on the generated time characteristic data ;
And an output unit for outputting the estimated intake amount. The estimation unit estimates one or more swallowing segments by dividing the biological signal, and estimates one intake amount of the ingested item for each swallowing segment,
The output unit outputs at least one of the bite intake amount and an aggregate value of the bite intake amount.
The prediction system according to claim 1. It is before Symbol first machine learning support vector classification,
The prediction system according to claim 1 or 2. Before Stories second machine learning support vector classification or a combination of a support vector classification and support vector regression,
The prediction system according to any one of claims 1 to 3. The biomedical signal is swallowing sound,
The prediction system as described in any one of Claims 1-4. The time characteristic data is an amplitude histogram,
The prediction system according to any one of claims 1 to 5. A prediction method performed by a prediction system comprising a processor, comprising:
An acquisition step of acquiring a biological signal generated by the swallowing movement of the subject;
The power spectral density is generated from the biological signal, the power spectral density is used as an explanatory variable , and the generated power spectral density is analyzed by first machine learning corresponding to classification of behavior using the throat. A determination step of determining whether the subject has ingested the intake;
Wherein when it is determined that the viewer has ingested, time time characteristics data indicating the characteristic form the biological signal or al production, second corresponding to the time characteristic data as explanatory variables and intake classification Estimating the intake of the intake of the subject by performing machine learning on the generated time characteristic data ;
And an output step of outputting the estimated intake amount. An acquisition unit for acquiring a biological signal generated by the swallowing motion of the subject;
The power spectral density is generated from the biological signal, the power spectral density is used as an explanatory variable , and the generated power spectral density is analyzed by first machine learning corresponding to classification of behavior using the throat. A determination unit that determines whether the subject has ingested an intake;
Wherein when it is determined that the viewer has ingested, time time characteristics data indicating the characteristic form the biological signal or al production, second corresponding to the time characteristic data as explanatory variables and intake classification An estimation unit configured to estimate the intake of the intake of the subject by performing machine learning on the generated time characteristic data ;
A prediction program for causing a computer to function as an output unit that outputs the estimated intake amount.