JP7162232B1 - Heart rate classification device and bather monitoring system - Google Patents

Abstract

Kind Code: A1 A heart rate classification device and a bathing person monitoring system capable of separating electrocardiographic beat signals corresponding to changes in heart rate intervals are provided. Kind Code: A1 A heartbeat classifying apparatus for estimating, based on an electrocardiographic signal, a belonging classification indicating a classification of each heartbeat of the electrocardiographic signal, wherein a part of the electrocardiographic signal is included from the electrocardiographic signal. extraction means for extracting a partial signal; probability calculation means for calculating a probability that the partial signal extracted by the extraction means is a reference partial signal containing a peak of an electrocardiographic signal; and probability calculated by the probability calculation means. separating means for separating an electrocardiographic beat signal representing one cycle of the electrocardiographic signal based on the partial signal having a threshold value or more; and an affiliation classification estimation means for estimation. [Selection drawing] Fig. 1

Description

The present invention relates to a heart rate classification device and a bather monitoring system.

Conventionally, a technique for classifying a user's heartbeat has attracted attention. Along with this, there is a need for a technique for separating an electrocardiographic beat signal representing an electrocardiographic signal for one cycle of the user's heartbeat for classifying the user's heartbeat. As a technique for separating an electrocardiogram beat signal, a technique such as that disclosed in Patent Document 1 has been proposed.

Japanese Patent Application Laid-Open No. 2004-200000 discloses a technique for separating an electrocardiographic beat signal based on partial signals extracted by a user-selected fixed-size extraction window.

WO2011/094105

However, in the case of non-deterministic signals such as electrocardiographic signals, the intervals between heartbeats change all the time. separation is required. However, in the technique for separating the electrocardiographic beat signal disclosed in Patent Document 1, the electrocardiographic beat signal is separated based on the partial signals extracted by the fixed-size extraction window selected by the user. Therefore, there is a problem that an electrocardiographic beat signal cannot be separated corresponding to changes in heartbeat intervals.

SUMMARY OF THE INVENTION Accordingly, the present invention has been devised in view of the above-described problems, and its object is to provide a heartbeat classification apparatus and a bathing person monitoring apparatus capable of separating electrocardiographic beat signals corresponding to changes in heartbeat intervals. It is to provide a system.

A heartbeat classification device according to a first aspect of the invention is a heartbeat classification device for estimating, based on an electrocardiographic signal, a classification indicating a classification of each heartbeat of the electrocardiographic signal, wherein a part of the electrocardiographic signal is extracted from the electrocardiographic signal. extraction means for extracting a partial signal containing the peak of the electrocardiographic signal; probability calculation means for calculating the probability between the partial signal extracted by the extraction means and the reference partial signal containing the peak of the electrocardiographic signal; Separating means for separating an electrocardiographic beat signal representing one cycle of the electrocardiographic signal based on the partial signal whose probability is equal to or greater than a threshold; and an affiliation classification estimation means for estimating the

A heart rate classification apparatus according to a second aspect of the invention is the first aspect of the invention, wherein the input data based on the reference partial signal and the probability associated with the input data are used as a pair of probability learning data, and a plurality of the probability learning data are used. and calculating the probability based on the partial signal extracted by the extraction means by referring to the probability model generated by machine learning using the extraction means.

A heartbeat classification apparatus according to a third invention is characterized in that, in the first invention, the extracting means extracts the partial signals based on heartbeat information about heartbeats.

A heart rate classification apparatus according to a fourth invention is characterized in that, in the first invention, the extracting means newly extracts partial signals based on the electrocardiographic beat signal separated by the separating means.

A heartbeat classification apparatus according to a fifth aspect of the invention is the first aspect, wherein the separating means calculates the length of one cycle of the electrocardiographic beat signal based on distances between a plurality of peaks in the electrocardiographic signal. It is characterized by

A heart beat classification apparatus according to a sixth invention is the heart beat classification apparatus according to any one of the first invention to the fifth invention, wherein the belonging classification estimating means converts the electrocardiographic beat signal separated by the separating means into a scalogram image, which is obtained in advance. Input data based on the training scalogram image and the affiliation classification linked to the input data as a pair of learning data, referring to a neural network model generated using a plurality of the learning data, and converting the scalogram It is characterized by estimating a classification belonging to an image as an input.

A bathing person monitoring system according to a seventh aspect of the present invention comprises a bathroom environment sensor that measures environmental data indicating any one or more of temperature, humidity, illuminance, reverberation, scent, and air pressure provided in a bathroom; bather information acquisition means for acquiring health data of a bather, one or more affiliation classifications estimated by the heart rate classification apparatus according to the sixth invention, environmental data measured by the bathroom environment sensor and the Based on bathing information obtained by combining health data obtained by the bathing person information obtaining means, referring to a state database recording state data indicating the state of the bathing person with respect to previously obtained belonging classification and bathing information, A state determining means for determining state data of a bather and treatment data indicating a method of treatment for state data obtained in advance and state data are used as a pair of learning data for determination, and a plurality of learning data for determination are recorded. and action determination means for determining action data corresponding to the state data determined by the state determination means by referring to the action information database.

A bathing person monitoring system according to an eighth aspect of the invention is characterized in that, in the seventh aspect of the invention, transmission means for transmitting treatment data determined by the treatment determination means to an external device, and transmitting the treatment data determined by the treatment determination means to the bather. presenting means for presenting;

A bathing person watching system according to a ninth aspect of the invention is characterized in that, in the seventh aspect of the invention, it further comprises drainage means for draining a bathtub provided in the bathroom based on the treatment data determined by the treatment determination means.

A bathing person watching system according to a tenth invention is, in the seventh invention, a treatment result acquisition means for acquiring result data indicating a treatment result for the treatment data determined by the treatment determination means; The state data determined by the state determination means and the treatment data determined by the treatment determination means according to the result data obtained are used as a pair of learning data for determination, and are newly recorded in the treatment information database. and means.

According to the first to tenth inventions, the separating means separates the electrocardiographic beat signal representing one cycle of the electrocardiographic signal based on the partial signal whose probability is equal to or higher than the threshold. Therefore, the electrocardiographic beat signal can be separated based on the partial signal that has a high probability of being the reference partial signal containing the peak of the electrocardiographic signal. As a result, it is possible to separate electrocardiographic beat signals corresponding to changes in heartbeat intervals.

In particular, according to the second invention, the probability calculation means refers to the probability model and calculates the probability based on the partial signal. Therefore, it is possible to calculate the probability with higher accuracy.

In particular, according to the third invention, the partial signal is extracted based on the heartbeat information regarding the heartbeat. This makes it possible to extract a partial signal having a window width suitable for each user, for example, based on the user's heart rate. As a result, it is possible to separate electrocardiographic beat signals with high accuracy.

In particular, according to the fourth invention, the extraction means newly extracts the partial signal based on the electrocardiographic beat signal. As a result, it is possible to extract partial signals at appropriate positions as the user's heartbeat changes. It is possible to prevent the extraction of partial signals.

In particular, according to the fifth invention, the separating means calculates the length of one cycle of the electrocardiographic beat signal based on the distances between the peaks. This makes it possible to separate electrocardiographic beat signals with an appropriate period even if, for example, the interval between heartbeats changes.

In particular, according to the sixth invention, the belonging class estimating means refers to the neural network model, receives the converted scalogram image as an input, and estimates the belonging class. Therefore, by using scalogram images as input data, the size of the CNN model can be reduced, and heart rate classification can be performed at low cost and at high speed.

According to the seventh to tenth inventions, the state data of the bather is determined by referring to the state database based on the belonging classification and the bathing information. This makes it possible to determine the state of the bather with high accuracy. Further, the treatment determination means determines treatment data for the status data determined by the status determination means. This makes it possible to determine with high accuracy the appropriate treatment for the bather.

In particular, according to the eighth invention, the treatment data is transmitted to the external device and the treatment data is presented to the bather. This makes it possible to present an appropriate treatment method to the bather and the bather's family or caregiver who owns the external terminal.

In particular, according to the ninth invention, the drainage means drains water from the bathtub provided in the bathroom based on the treatment data. As a result, it is possible to automatically drain the water according to the state of the bather.

In particular, according to the tenth invention, the recording means makes the state data and the treatment data a pair of learning data for judgment according to the result data, and newly records them in the treatment information database. Thereby, appropriate learning data can be fed back according to the result.

FIG. 1 is a schematic diagram showing an example of a bathing person watching system according to the present embodiment. FIG. 2 is a schematic diagram showing an example of a bathtub in this embodiment. FIG. 3A is a schematic diagram showing an example of the configuration of the heartbeat classification device according to this embodiment. FIG. 3(b) is a schematic diagram showing an example of the function of the heart rate classification device according to this embodiment. FIG. 4 is a flow chart showing an example of the operation of the heartbeat classification device according to this embodiment. FIG. 5(a) is a schematic diagram showing an example of an electrocardiographic signal. FIG. 5B is a schematic diagram showing three examples of electrocardiographic beat signals. FIG. 5(c) is a schematic diagram showing three examples of scalogram images corresponding to (b). FIG. 6 is a schematic diagram showing seven examples of belonging classifications. FIG. 6 shows the electrocardiographic signals on the left and the corresponding scalogram images on the right. FIG. 7 is a schematic diagram showing an example of a neural network model used for classification of electrocardiographic beat signals. FIG. 8 is a flowchart showing an example of an operation for extracting an electrocardiographic beat signal from an electrocardiographic signal. FIG. 9 is a schematic diagram illustrating the adaptive process of window movement for extracting partial signals from an electrocardiographic signal and determining the boundaries of the electrocardiographic beat signal. FIG. 10 is a flow chart illustrating an example of window parameter calculation. FIG. 11 is a schematic diagram showing an example of an electrocardiographic signal triplet for generating a reference partial signal, which is learning data for a stochastic model. FIG. 4 is a schematic diagram showing how an ECG triplet segment and critical cpl, cpm, cpr are used to generate 12 reference sub-signals that are assigned a probability of 1; FIG. 13 is a schematic diagram showing how an ECG triplet segment and critical cpl, cpm, cpr are used to generate 12 reference sub-signals that are assigned a probability of 0; FIG. 14 is a schematic diagram showing an example of a probability evaluation model. FIG. 15 is a flow chart showing an example of the operation of the bathing person watching system according to this embodiment.

An example of a bathing person watching system according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic diagram showing an example of a bather monitoring system 100 according to this embodiment.

The bather watching system 100 is used to manage the condition of the bather.

The bather monitoring system 100 includes, for example, a heart rate classification device 1 connected via a communication network 4, a terminal 2, a server 3, and a bathroom 5, as shown in FIG.

The terminal 2 is owned by, for example, a bather's family or caregiver whose condition is managed by the bather monitoring system 100 , and is connected to the heart rate classification device 1 via the communication network 4 . As the terminal 2, for example, an electronic device such as a personal computer or a tablet terminal is used. The terminal 2 may have at least part of the functions of the heart rate classification device 1, for example.

The server 3 is connected to the heartbeat classification device 1 and the like via a communication network 4 . The server 3 stores various past data and the like, and various data are transmitted from the heart rate classification device 1 and the bathroom 5 as necessary. The server 3 may have, for example, at least part of the functions of the heartbeat classification device 1, and may perform at least part of the processing instead of the heartbeat classification device 1, for example.

The communication network 4 is, for example, an Internet network or the like to which the heart rate classification device 1 is connected via a communication circuit. The communication network 4 may be composed of a so-called optical fiber communication network. Moreover, the communication network 4 may be realized by a known communication technology such as a wireless communication network in addition to the wired communication network. Also, the communication network 4 may be a LAN (Local Area Network).

The bathroom 5 is a closed space for a bather to take a bath. The bathroom 5, as shown in FIG. and an electrode 55 . Also, the bathroom 5 may be provided with two or more electrodes 55 . Moreover, the bathroom 5 may be equipped with an air conditioner, lighting equipment, audio equipment, fragrance generator, etc., which are not shown. Sensor 53 , ventilation fan 56 , drainage device 52 , monitor 54 and electrodes 55 are also connected to heart rate classification device 1 via communication network 4 .

The drainage device 52 is a device for draining the inside of the bathtub 51 according to a command from the heart rate classification device 1 . The drainage device 52 is, for example, a device having an arbitrary drainage function that can be retrofitted, but is not limited to this, and any device may be used. The drainage device 52 may be, for example, a device that drains the inside of the bathtub 51 by opening the faucet of the bathtub 51 using a spring or the like according to a command from the heart rate classification device 1 .

The sensor 53 is a sensor that measures environmental data indicating any one or more of temperature, humidity, atmospheric pressure, illuminance, reverberation (music, etc.), and fragrance in the bathroom 5, and is a temperature sensor, a humidity sensor, and an atmospheric pressure sensor. etc. are used. The sensor 53 transmits environmental data to various devices via the communication network 4 .

The monitor 54 is a monitor for presenting treatment data indicating a treatment method for the bather in accordance with a command from the heart rate classification device 1 .

The electrodes 55 are electrodes for measuring the bather's electrocardiographic signal. The electrodes 55 measure electrocardiographic signals of the bather by measuring potential differences at a plurality of locations near the limbs, chest, etc. of the bather. A plurality of electrodes 55 may be embedded in the bathtub 51 . Alternatively, the electrodes 55 may be attached to the wall of the bathtub 51 . The bather can measure the electrocardiographic waveform without touching the electrodes 55 .

As the heartbeat classification device 1, an electronic device such as a laptop (notebook) PC or a desktop PC is used. For example, as shown in FIG. It has a unit 104 and I/Fs 105-107. Each configuration 101 - 107 is connected by an internal bus 110 .

The CPU 101 controls the heart rate classification device 1 as a whole. ROM 102 stores the operation code of CPU 101 . A RAM 103 is a work area used when the CPU 101 operates. The storage unit 104 stores various types of information such as databases and data to be learned. As the storage unit 104, for example, a data storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive) is used. For example, the heart rate classification device 1 may have a GPU (Graphics Processing Unit) (not shown).

The I/F 105 is an interface for transmitting/receiving various information to/from the terminal 2, the server 3, a website, etc. via the communication network 4 as necessary. The I/F 106 is an interface for transmitting/receiving information to/from the input unit 108 . A keyboard, for example, is used as the input unit 108 , and the user of the heartbeat classification device 1 or the like inputs various information, control commands for the heartbeat classification device 1 , or the like via the input unit 108 . The I/F 107 is an interface for transmitting and receiving various information to and from the display unit 109 . The display unit 109 displays various types of information or content stored in the storage unit 104 . A display is used as the display unit 109 , and is provided integrally with the input unit 108 in the case of a touch panel type, for example. A speaker may be used for the display unit 109 .

FIG. 3(b) is a schematic diagram showing an example of the functions of the heart rate classification device 1. As shown in FIG. The heartbeat classification device 1 includes an acquisition unit 11, a conversion unit 12, an affiliation classification estimation unit 13, a storage unit 14, an output unit 15, a treatment determination unit 16, a state determination unit 17, an extraction unit 18, and a separation unit. a part 19; Note that each function shown in FIG. 3B is realized by the CPU 101 using the RAM 103 as a work area and executing a program stored in the storage unit 104 or the like, and may be controlled by, for example, artificial intelligence. .

Acquisition unit 11 acquires various data such as environmental data, electrocardiographic signals, health data of bathers, and the like. Acquisition unit 11 acquires, for example, health data input from input unit 108, and may also acquire environmental data and electrocardiogram signals from bathroom 5 or the like via communication network 4, for example.

The extraction unit 18 extracts a partial signal including part of the electrocardiogram signal from the electrocardiogram signal acquired by the acquisition unit 11 .

The separating unit 19 calculates the probability that the partial signal extracted by the extracting unit 18 is the reference partial signal containing the critical point of the electrocardiographic signal. Further, the separating unit 19 separates an electrocardiographic beat signal representing one cycle of the electrocardiographic signal based on the partial signals for which the calculated probability is equal to or higher than the threshold.

The converter 12 converts the electrocardiogram beat signal separated by the separator 19 into a scalogram image.

The affiliation classification estimation unit 13 sets a pair of input data based on a learning scalogram image acquired in advance and an affiliation classification linked to the input data as learning data, and generates a neural network model using a plurality of learning data. , and the scalogram image converted by the conversion unit 12 is used as an input to estimate the affiliation class.

The state determination unit 17 performs bathing for the pre-acquired affiliation classification and bathing information based on the one or more affiliation classifications estimated by the affiliation classification estimation unit 13 and the bathing information obtained by combining the environmental data and the health data. The state data of the bather is determined by referring to a state database in which the state data indicating the state of the bather is recorded.

The treatment determination unit 16 sets state data obtained in advance and treatment data indicating a method of treatment for the state data as a pair of learning data for determination, and refers to a treatment information database in which a plurality of learning data for determination are recorded. Then, treatment data corresponding to the state data determined by the state determination unit 17 is determined.

The storage unit 14 retrieves various data such as databases stored in the storage unit 104 as necessary. The storage unit 14 stores various data acquired or generated by each configuration in the storage unit 104 as necessary.

The output unit 15 outputs various data. The output unit 15 outputs various data to the display unit 109 via the I/F 107, and also outputs various data to a plurality of terminals 2 or the like via the I/F 105, for example.

Next, an example of the operation of the bathing person watching system 100 according to this embodiment will be described.

The bather monitoring system 100 is executed via a program installed in the heart rate classification device 1, for example. That is, the state of the bather is managed through a program installed in the heart rate classification device 1 .

FIG. 4 is a flow chart showing an example of the operation of the heart rate classification device 1 according to this embodiment. In the operation of the bather monitoring system 100, the belonging classification is first estimated by each step shown in FIG.

First, in step S110, the bather's electrocardiogram signal is measured. For example, the electrodes 55 in the bathroom 5 measure the bather's electrocardiographic signal. In such a case, the electrodes 55 measure the electrocardiogram signal over the entire bathing time or a preset time. Electrodes 55 measure the bather's electrocardiogram signal for, for example, 10 minutes. Alternatively, the bather's bathing and bathing timings are automatically detected, and the electrocardiogram signal of the bather is measured over the entire bathing time from bathing to bathing. Bathroom 5 transmits the measured data to heart rate classification device 1 via communication network 4 .

As shown in FIG. 5(a), the electrocardiographic signal is a signal that indicates changes in cardiac potential (mV order) with respect to time (s), and includes a P wave that indicates atrial excitation, a QRS wave that indicates ventricular excitation, It includes the T wave, etc., in which ventricular excitation subsides. Also, one cycle of the electrocardiographic signal is the period until these specific signals are observed again, and indicates, for example, the period from the P wave to the next P wave. Also, one cycle of the electrocardiographic signal may be an electrocardiographic signal that is segmented at regular time intervals.

In step S110, the electrocardiographic signal is separated into electrocardiographic beat signals representing one cycle of the electrocardiographic signal. The electrocardiographic beat signal contains the critical points of the electrocardiographic signal. A critical point of an electrocardiographic signal is, for example, a QRS wave, but is not limited to this, and may be a peak, critical point, or extremum indicating the maximum value or minimum value of another signal. A critical point of an electrocardiogram signal may be determined by the magnitude of the absolute value with respect to the threshold. The electrocardiographic beat signal includes one QRS wave, for example, as shown in FIG. 5(b). The details of the method of separating the electrocardiographic beat signal in step S110 will be described later.

Next, in step S120, the electrocardiographic beat signal is converted into a scalogram image. A scalogram image is an image showing a scalogram of an electrocardiographic beat signal, as shown in FIG. 5(c). In step S160, for example, the conversion unit 12 may convert the electrocardiographic beat signal into a scalogram image or a spectrumgram image.

Next, in step S130, the scalogram image is used as an input to estimate the belonging class. In such a case, the affiliation classification estimation unit 13 uses a pair of input data based on a learning scalogram image acquired in advance and an affiliation classification linked to the input data as learning data, and generates using a plurality of learning data A neural network model is referred to, and a scalogram image converted by a conversion process is used as an input to estimate belonging classification. In addition, in estimating the classification, it is not necessary to use the methods described in steps S120 and S130 above, and any method may be used to estimate the classification of the electrocardiographic beat signals separated in step S110. good.

Affiliation classification is data indicating the heartbeat classification of the electrocardiogram signal. The affiliation classification is data indicating the classification as shown in Table 1, for example.

ニューラルネットワークモデルは、例えば複数の学習用データを用いた機械学習により生成された、学習済みモデルを含んでもよい。学習済みモデルは、例えばＣＮＮ（Convolutional Neural Network）等のニューラルネットワークモデルを示すほか、ＳＶＭ（Support Vector Machine）等を示す。また、機械学習として、例えば深層学習を用いることができる。図６は、学習用データの一例を示す模式図である。学習用データは、例えば図６に示すように、学習用スカログラム画像と、所属分類とが紐づけられる。

A neural network model may include a trained model generated by machine learning using a plurality of training data, for example. A trained model indicates, for example, a neural network model such as a CNN (Convolutional Neural Network), or an SVM (Support Vector Machine). As machine learning, for example, deep learning can be used. FIG. 6 is a schematic diagram showing an example of learning data. In the learning data, for example, as shown in FIG. 6, the learning scalogram image and the affiliation classification are linked.

The neural network model stores associations with degrees of association, for example, between scalogram images or electrocardiographic beat signals (input data) and belonging classes (output data). The degree of association indicates the degree of connection between the input data and the output data. For example, it can be determined that the higher the degree of association, the stronger the connection between the data. The degree of association may be indicated by, for example, a percentage or the like, and may be indicated by several levels.

For example, associations are constructed by the degree of connectivity between many-to-many information (multiple input data versus multiple output data). Associations are updated as appropriate during the course of machine learning, and represent optimized functions (classifiers), for example, based on multiple input data and multiple output data. Note that the relevance may have, for example, a plurality of relevance degrees indicating the degree of connection between each piece of data. The degree of association can correspond to a weight variable, for example when the database is built with neural networks.

For this reason, the heart rate classification device 1 selects output data suitable for input data, for example, by using association based on all the results determined by the classifier. This makes it possible to quantitatively select output data suitable for the input data not only when the input data is the same as or similar to the learning input data, but also when they are dissimilar.

Relevance may indicate, for example, the degree of connection between a plurality of output data and a plurality of input data. In this case, by using the association, for each of the plurality of output data (“affiliation class A” to “affiliation class C”), the plurality of input data (“scalogram image A” to “scalogram image C”) The degree of relationship can be linked and stored. For this reason, it is possible to associate a plurality of output data with a specific degree of association with one input data, for example, via association. This makes it possible to obtain the belonging classification for the scalogram image.

Relevance has, for example, a plurality of degrees of relevance linking each output data and each input data. The degree of association is indicated, for example, in three or more levels such as percentage, 10 levels, or 5 levels, and is indicated, for example, by line characteristics (such as thickness). For example, "scalogram image A" included in the input data indicates the degree of association AA "73%" between "belonging class A" included in the output data, and "belonging class B" included in the output data. It shows the degree of association AB “12%” between and the degree of association AC “15%” with “belonging class C” included in the output data. That is, the "relevance degree" indicates the degree of connection between each piece of data. For example, the higher the degree of association, the stronger the connection between each piece of data.

Further, the neural network model may be provided with at least one or more hidden layers between the input data and the output data for machine learning. The degree of association described above is set in either or both of the input data and the hidden layer data, and this serves as weighting for each data, and output selection is performed based on this. Then, when the degree of association exceeds a certain threshold, the output may be selected.

In addition, when classifying the separated electrocardiographic beat signals in step S110 without using the methods of steps S120 and S130 described above, for example, a neural network generated using learning data as shown in FIG. , the classification may be estimated based on the electrocardiographic beat signals separated by step S110. In such a case, the learning data may have the reference electrocardiographic beat signal as the input and the belonging classification as the output. The reference electrocardiographic beat signal is an electrocardiographic beat signal previously acquired for use as learning data. In this case, by using the association, a plurality of input data (“electrocardiographic beat signal A” to “electrocardiographic beat signal C”) can be linked and stored.

The affiliation classification estimating unit 13 receives the scalogram image converted in step S160 using the neural network model described above, and outputs the affiliation classification.

In addition, the affiliation classification estimation unit 13 refers to the neural network model generated using the learning data having the plurality of learning scalogram images as input data, and receives the plurality of scalogram images converted in step S160 as input, Classification may be inferred. This neural network model differs from the neural network model described above in that the input data includes multiple scalogram images.

This input data includes two or more scalogram images. Also, the input data may include a plurality of scalogram images that are continuous in time series. As a result, for example, by inputting a plurality of scalogram images that are continuous in time series, it is possible to determine the characteristics from the pattern of temporal changes in the electrocardiogram. This makes it possible to improve the accuracy when estimating the belonging classification. The input data may also include a plurality of scalogram images based on a plurality of electrocardiographic signals acquired using a plurality of channels divided for each measurement site by the electrodes 55 .

Further, in step S130, the affiliation class estimation unit 13 receives a plurality of scalogram images as input, estimates a plurality of estimation affiliation classes, and estimates the affiliation class based on the estimated plurality of estimation affiliation classes. good. The affiliation classification for estimation is an affiliation classification for estimating one affiliation classification from a plurality of affiliation classifications. In this case, the affiliation classification estimating unit 13 estimates a plurality of estimation affiliation classifications from a plurality of scalogram images based on a plurality of electrocardiographic signals acquired using a plurality of channels divided for each measurement site by the electrode 55, for example. However, the largest number of affiliation classifications may be selected from a plurality of estimation affiliation classifications. This makes it possible to comprehensively estimate the belonging class from, for example, a plurality of scalogram images acquired using a plurality of channels. Therefore, it is possible to estimate the affiliation classification with high accuracy.

In step S130, the classification affiliation estimating unit 13 receives one scalogram image as an input, estimates a plurality of classifications for estimation, and estimates the classification based on the estimated plurality of classifications for estimation. good. In such a case, for example, using an arbitrary trained model whose input is a scalogram image and whose output is an affiliation classification for estimation, multiple An inferred affiliation class may be inferred.

Further, in step S130, the affiliation classification estimation unit 13 calculates the degree of similarity between the plurality of scalogram images converted by the conversion unit 12 and the pre-obtained reference scalogram image. In this case, it is possible to estimate the affiliation classification.

The reference scalogram image may be, for example, a scalogram image based on the bather's electrocardiogram signal to be estimated. By comparing the degree of similarity with this reference scalogram image, it can be estimated in step S110 whether or not the measured electrocardiographic signal is a scalogram image based on the electrocardiographic signal of the bather to be estimated. can. Therefore, the estimation can be performed automatically only when the bather to be judged is bathing, so that the estimation can be performed at low cost and at high speed.

Also, the reference scalogram image may be, for example, a scalogram image based on an electrocardiographic signal when the bather to be estimated is taking a bath. By comparing the degree of similarity with this reference scalogram image, it is determined in step S110 whether or not the measured electrocardiographic signal is a scalogram image based on the electrocardiographic signal when the bather to be estimated is bathing. can be estimated. Therefore, the estimation can be performed automatically only when the bather to be judged is bathing, so that the estimation can be performed at low cost and at high speed.

By performing each step described above, the operation of estimating the affiliation classification of the bathing person watching system 100 is completed. This allows the size of the CNN model to be reduced by using scalogram images as input data, enabling low-cost and high-speed heart rate classification.

Next, details of the operation of extracting an electrocardiographic beat signal from the electrocardiographic signal in step S110 will be described. FIG. 8 is a flowchart showing an example of an operation for extracting an electrocardiographic beat signal from an electrocardiographic signal.

First, in step S20, the extraction unit 18 extracts partial signals from the electrocardiographic signal. FIG. 9 is a schematic diagram illustrating the window adaptation process for extracting partial signals from an electrocardiographic signal and determining the boundaries of the electrocardiographic beat signal. First, in step S20, the extraction unit 18 extracts the electrocardiogram signal included in the window of width w _n as a partial signal.

Next, in step S21, the extraction unit 18 recalculates and updates the parameters of the window. The parameters of the window are, for example, window width w _n , window step width s _n , maximum estimated future RR interval (maxRRI), minimum estimated future RR interval (minRRI), past cp (cp _n-2 , cp _n-1 ), current cp _n , small step forward (fwd), window position (wposition), offset value, and so on. w _n indicates the width of the window, e.g., the average of RR intervals calculated from a plurality of predetermined (e.g., 16 beats) RR intervals extracted from previously acquired electrocardiographic signals, or the user's average It is calculated from the heart rate (for example, using the latest 16 heart beats in the past). The step width s _n indicates the distance the window moves each time a partial signal is extracted. The maximum value of the RR interval (maxRRI) indicates the expected maximum RRI limit, for example, the product of the average value of the RR interval calculated from a plurality of partial signals extracted from the previously acquired electrocardiographic signal and a certain constant. Calculated. Also, the maximum value of the RR interval (maxRRI) may be updated each time the electrocardiographic beat signal is separated. The minimum value of the RR interval (minRRI) indicates the expected minimum RRI limit, for example, the average of the RR intervals calculated from a plurality of (eg, 16 beats) RR intervals extracted from the electrocardiographic signal acquired in the past. Calculated by multiplying with a constant. Also, the minimum value of the RR interval (minRRI) may be updated each time the electrocardiographic beat signal is separated. Small step forward (fwd) is a parameter for finely moving the window as necessary. The offset value indicates the starting position of the next window and is used to exclude critical points of the ECG signal already contained in the currently detected partial signal. This makes it possible to minimize the number of steps for detecting the next partial signal. The offset value is calculated using the new auto-adapted window width, the current cp, and a preset constant. The RR interval is the distance between adjacent critical points in the electrocardiogram signal.

Next, a step S22 evaluates whether the electrocardiographic beat signal calculated in S21 is valid or invalid based on the state of the partial signal. If the evaluation result is a valid ECG beat signal, the process moves to step S23. If the evaluation result is an invalid ECG beat signal, the process moves to step S35. This evaluation result may be determined, for example, by comparing the evaluation value with a preset threshold, using the probability that the partial signal is the reference partial signal containing the critical point of the electrocardiogram signal as an evaluation value. For example, if the probability is greater than 0.9, the partial signal may be a valid ECG beat signal, and if the probability is less than 0.9, the partial signal may be an invalid ECG beat signal.

In step S22, the extraction unit 18 calculates, for example, the probability that the partial signal is the reference partial signal containing the critical point of the electrocardiographic signal. A reference partial signal is a partial signal containing a critical point of an electrocardiographic signal acquired in advance. FIG. 8 processes a series of partial signals while sliding an adaptive window to extract valid electrocardiographic beat signals. FIG. 10 is used further in FIG. 8 after calculating the window parameters and calculating the similarity between the partial signal and the reference partial signal.

In step S<b>23 , the extraction unit 18 determines whether the number of partial signals whose probability evaluated in step S<b>22 is equal to or greater than the threshold is greater than 16. If the number is greater than 16, the process proceeds to step S24, and if the number is 16 or less, the process proceeds to step S35. Thus, initial setting of window parameters is determined based on the 16 or more partial signals detected in step S23. Also, the value of 16 is an example, and any value may be used.

In step S35, the extraction unit 18 moves the window to extract the next partial signal, and extracts the partial signal again in step S20. In such a case, for example, the window is moved by the step width s _n of the window from the window position of the window parameter updated in step S21.

In step S36, the extraction unit 18 moves the window to the side of the critical point of the partial signal evaluated in step S22, and again extracts the partial signal in step S20. In such cases, the offset value is calculated based on the critical point. The extraction unit 18 may use, for example, the sum of the critical point and the constant as the offset value. As a result, the next partial signal can be extracted without moving the window to an unnecessary location before the critical point, so that the partial signal can be extracted more efficiently and the partial signal containing the same critical point can be extracted. can be prevented.

Further, in step S37, the initial settings of the window parameters may be estimated instead of determining the initial settings of the window parameters in steps S20 to S23.

Next, in step S24, the extraction unit 18 extracts partial signals for actually separating the electrocardiographic beat signal based on the heartbeat information regarding the heartbeat. Heartbeat information is, for example, a window parameter. Also, the heartbeat information may be biological information such as the user's heartbeat rate. In step S24, the extraction unit 18 extracts partial signals for actually separating the electrocardiographic beat signal, for example, using the initial setting of the window parameters determined in step S23. In step S24, the partial signal may be extracted using the window parameter estimated in step S37.

Next, in step S25, the window parameters are recalculated and updated.

Next, in step S26, the partial signals extracted in step S24 are evaluated. In step S26, if the probability of the partial signal extracted in step S24 is equal to or greater than the threshold, the process proceeds to step S27 or step S30, and if the probability is equal to or less than the threshold, the process proceeds to step S38. If the probability is equal to or greater than the threshold and the output pattern B of the electrocardiographic beat signal is used in step S26, the process proceeds to step S27. If the probability is equal to or greater than the threshold and the output pattern A of the electrocardiographic beat signal is used, the process proceeds to step S30. You may The output pattern of this electrocardiographic beat signal may be changed as required. Alternatively, a plurality of output patterns of the electrocardiographic beat signal may be used to output each electrocardiographic beat signal. Accordingly, by classifying the output electrocardiographic beat signals using a plurality of output patterns, it is possible to further improve the classification accuracy. Further, in step S26, a partial signal whose evaluation value is equal to or greater than a threshold value may be separated as an electrocardiogram beat signal.

In step S38, the extraction unit 18 moves the window to extract the next partial signal, and extracts the partial signal again in step S24. In such a case, the extraction unit 18 moves the window by the step width s _n of the window from the window position of the parameter of the window updated in step S25, for example.

When the output pattern B of the electrocardiographic beat signal is used, in step S27, the extraction unit 18 rearranges the clipping points based on the window parameters updated in step S25. In such a case, the extraction unit 18 determines the start position and end position of the clipping point based on, for example, the critical point, the RR interval, the average value of past RR intervals, and the like.

In step S28, the extraction unit 18 extracts an electrocardiogram beat signal based on the start position and end position of the critical points rearranged in step S27. Alternatively, the process may proceed to step S38 in order to extract the next partial signal.

In step S29, the extraction unit 18 predicts the probability of the extracted electrocardiographic beat signal. The extraction unit 18 calculates, for example, the probability of the electrocardiographic beat signal and the reference partial signal including the critical point of the electrocardiographic signal as an evaluation value. The extraction unit 18 outputs the extracted electrocardiographic beat signal if the probability is equal to or greater than the threshold.

When the output pattern A of the electrocardiographic beat signal is used, in step S30, the extraction unit 18 newly extracts a partial signal based on the separated electrocardiographic beat signal. The extraction unit 18 calculates the start position of the clipping point for extracting a new partial signal, for example, based on the window parameters updated in step S25. In such a case, the extractor 18 may, for example, calculate the starting position from past critical points and one or two preset constants. Further, the extraction unit 18 may calculate the start position of the clipping point for newly extracting the partial signal based on the output electrocardiogram beat signal, for example, in step S26 or S29. The extraction unit 18 calculates the start position of the clipping point, for example, based on the critical point and the offset value of the electrocardiographic beat signal separated in the past.

In step S31, the extraction unit 18 determines whether the start position calculated in step S30 is smaller than the critical point. If it is smaller than the critical point, the extraction unit 18 proceeds to step S32. If it is equal to or greater than the critical point, the extraction unit 18 proceeds to step S33.

In step S32, the extraction unit 18 adjusts the start position calculated in step S30. In such a case, the extractor 18 adjusts the start position based on the critical point.

In step S33, the extraction unit 18 calculates the end position of the critical point. The extraction unit 18 calculates the end position, for example, based on the RR interval and the start position.

In step S34, the separation unit 19 performs electrocardiogram based on the start position of the clipping point calculated in step S30, or the start position of the clipping point adjusted in step S32 and the end position of the clipping point calculated in step S33. Extract the beat signal. Moreover, the separating unit 19 may output the extracted electrocardiographic beat signal. Also, the extraction unit 18 may proceed to step S38 in order to extract the next partial signal.

By performing the steps described above, the operation of extracting the electrocardiographic beat signal from the electrocardiographic signal is completed. Thus, the electrocardiographic beat signal can be separated based on the partial signal that has a high probability of being the reference partial signal containing the critical point of the electrocardiographic signal. Therefore, it is possible to separate electrocardiographic beat signals corresponding to changes in heartbeat intervals. Also, by using the output pattern A of the electrocardiographic beat signal, the calculation of the window boundary is performed based on the fitted window, the position of the latest critical point, and a predetermined constant. That is, boundary calculations are calculated with reference to the latest critical point without considering neighboring critical points. This allows real-time electrocardiographic beat signal separation. Also, by using the output pattern B of the ECG beat signal, the calculation of the window boundary is calculated based on the adjacent critical points. This allows for more accurate window bounds calculations.

Next, a method for calculating window parameters will be described. FIG. 10 is a diagram showing the flow of window parameter calculation.

First, in step S1, the extractor 18 removes noise from the extracted partial signals and normalizes them. In such a case, the QRS complex may be aligned with the center of the electrocardiographic beat signal. At this time, the electrocardiogram beat signal may include the electrocardiogram signals from before to after a certain time from the center of the critical point of the electrocardiogram beat signal. Also, the critical point does not necessarily have to be the center, and the center may be aligned at a fixed position.

Next, in step S2, the extraction unit 18 centers the critical point of the partial signal subjected to noise removal and normalization in step S1. Noise removal may be performed using any filter. Normalization may be performed by setting the maximum value of the electrocardiographic beat signal to 1 and the minimum value to 0, for example.

Next, in step S3, the extraction unit 18 calculates the probability that the partial signal centered on the critical point in step S2 is the reference partial signal including the critical point of the electrocardiogram signal. A reference partial signal is a partial signal containing a critical point of an electrocardiographic signal acquired in advance. Also, the reference partial signal may be a partial signal in which noise removal and normalization are performed in the same manner as in steps S1 and S2 and the critical point is centered.

Further, in step S3, the extraction unit 18 sets the input data based on the reference partial signal and the probability associated with the input data as a pair of probability learning data, and generates the data by machine learning using a plurality of probability learning data. The probabilistic model may be referred to, and the probability based on the partial signal centered on the critical point may be calculated in step S2.

FIG. 11 is a schematic diagram showing an example of an electrocardiographic signal triplet for generating a reference partial signal, which is learning data for a stochastic model. The reference partial signal may be obtained by processing for increasing the number of samples of the learning data set, for example, using an electrocardiographic signal triplet as shown in FIG. With ECG triplet segments and critical cpl, cpm, cpr, 12 reference sub-signals assigned a probability of 1 (1 typical standard valid reference sub-signal + 11 alternative equivalent valid 1 is a schematic diagram showing a method of generating a partial signal). FIG. FIG. 13 is a schematic diagram showing how ECG triplet segments and critical cpl, cpm, cpr are used to generate 12 reference sub-signals (invalid reference sub-signals) that are assigned a probability of 0; In FIG. 13, sl indicates the left edge of the selected electrocardiographic signal triplet, and sr indicates the right edge of the selected electrocardiographic signal triplet. d1 and d2 are the RR intervals between critical points of the ECG signal. As shown in FIG. 13, the triplet of the electrocardiographic signal includes information such as critical points (cp) and clipping points. cp is a characteristic point such as a critical point of the electrocardiographic signal, and its specific position depends on the form of the electrocardiographic signal. Most often, it is the critical point location of the QRS complex, but not necessarily the critical point. cpm is the middle cp, cpl is the left cp of cpm, and cpr is the right cp of cpm. bl and br indicate the coordinate positions of both ends of the partial signal. As shown in FIG. 13, it is possible to automatically extract a partial signal based on the values of the RR interval, d1, and d2, and an arbitrary preset constant. An invalid electrocardiographic beat signal of 0 may be used as learning data. This reference partial signal in FIG. 12 is an example of a triplet of each electrocardiographic signal for generating a set that extends the training data. For example, bl indicates the left edge of the partial signal and br indicates the right edge of the partial signal. d1 and d2 are used to calculate bl and br based on preset centering ratios. As a result, the generated partial signal may be used as learning data as a valid reference partial signal with a probability of 1. The above critical points (cp) and clipping points are reference points for performing left and right shifts, and expansion procedures and other manipulations to increase the sample size, as described in FIGS. 12 and 13. may be used as The signal thus extracted is used as a reference partial signal for learning of the probability model. This enables automatic extension of the reference partial signal.

During the segmentation process, two clipping points are determined based on the newly calculated window length (updated window parameters), the critical point, and the preset center ratio. These two clipping points are used to extract the ECG beat signal after a valid signal segment is detected. As such, the main ECG signal feature (such as a normal QRS or an abnormal QRS-like waveform) is placed approximately in the middle of the split ECG beat signal.

A probabilistic model may include a trained model generated by machine learning using a plurality of training data, for example. A trained model indicates, for example, a neural network model such as a CNN (Convolutional Neural Network), or an SVM (Support Vector Machine). The trained model may be trained using, for example, k-fold cross validation. As machine learning, for example, deep learning can be used. Also, the probability model may be a model generated by unsupervised learning using a plurality of reference partial signals. In the learning data, for example, reference partial signals and probabilities are linked.

The probabilistic model stores, for example, relationships having degrees of association between reference partial signals (input data) and probabilities (output data). For this reason, the heart rate classification device 1 selects output data suitable for input data, for example, by using association based on all the results determined by the classifier. This makes it possible to quantitatively select output data suitable for the input data not only when the input data is the same as or similar to the learning input data, but also when they are dissimilar.

Relevance may indicate the degree of connection between, for example, a plurality of output data and a plurality of input data. In this case, the degree of association A1 indicates the degree of association with valid reference partial signals, and the degree of association A0 indicates the degree of association with invalid electrocardiographic beat signals. In this case, by using associativity, it is possible to link the degree of relationship between the plurality of input data to each of the plurality of output data and store them. For this reason, a plurality of input data can be linked to one output data, for example, via association. Thereby, the probability can be obtained for the reference partial signal. Also, the probabilities used as output data may be, for example, only two types of probabilities 1 and 0, but the present invention is not limited to this, and probabilities of 3 or more set to arbitrary values may be used.

Moreover, the probability model may be machine-learned by providing at least one or more hidden layers between the input data and the output data. The degree of association described above is set in either or both of the input data and the hidden layer data, and this serves as weighting for each data, and output selection is performed based on this. Then, when the degree of association exceeds a certain threshold, the output may be selected. As shown in FIG. 14, a partial signal is input to this probability model, and the output probability is compared with a threshold to determine whether the partial signal is a valid electrocardiographic beat signal or an invalid electrocardiographic beat signal. becomes possible.

Further, invalid electrocardiographic beat signals with probability 0 that do not contain critical points may be used as reference partial signals as learning data for the stochastic model. In such a case, a partial signal that does not include a critical point as shown in FIG. 13 may be used as a reference partial signal. In this case, a probability model may be used to calculate the degree of divergence from the reference partial signal.

The extracting unit 18 uses the above-described probability model, receives as input the partial signal arranged at the center of the critical point in step S2, and outputs the probability.

Next, in step S4, the extraction unit 18 determines whether the probability calculated in step S3 exceeds a threshold. If the probability exceeds the threshold, the extraction unit 18 proceeds to step S5. If the probability is equal to or less than the threshold, the extraction unit 18 proceeds to step S9.

Next, in step S5, the extraction unit 18 calculates cp of the partial signal whose probability has been calculated in step S3. The extraction unit 18 may set, for example, the point at which the absolute value of the difference from the threshold in the partial signal is the largest as cp.

Next, in step S6, the extraction unit 18 compares the cp calculated in step S5 with the RR interval, which is the distance between the cp and the adjacent cp, and the expected minimum value of the RR interval. If the RR interval is greater than the expected minimum value of the RR interval, the extraction unit 18 proceeds to step S10. If the RR interval is equal to or less than the expected minimum value of the RR interval, the extraction unit 18 proceeds to step S7.

In step S7, the extraction unit 18 compares the cp calculated in step S5 with the RR interval, which is the distance between the cp and the adjacent cp, and the expected maximum value of the RR interval. If the RR interval is greater than the expected maximum value of the RR interval, the extraction unit 18 proceeds to step S8. If the RR interval is equal to or less than the expected maximum value of the RR interval, the extraction unit 18 proceeds to step S11.

In step S8, the extraction unit 18 updates the cp of the window parameter calculated in step S5. Also, in step S8, the past cp may be updated sequentially.

In step S9, the extractor 18 calculates the position of the next window. In step S9, the extraction unit 18 updates the window parameter to the calculated window position.

In step S10, the extractor 18 calculates an offset value. The extraction unit 18 calculates the offset value based on, for example, the sum of the critical point and fwd. Also, the extraction unit 18 may update the parameter of the window to the calculated offset value.

In step S11, the extraction unit 18 updates the parameters of the window. In step S11, for example, the RR interval, expected maximum and expected minimum values of the RR interval, window width w _n and step width s _n are calculated.

In step S12, the extraction unit 18 resets fwd.

In step S13, the extraction unit 18 updates the cp of the window parameter calculated in step S5. Also, in step S13, the past cp may be updated sequentially.

In step S14, the extractor 18 calculates an offset value. The extractor 18 calculates the offset value based on the critical point, for example. Also, the parameters of the window may be updated to the calculated offset value.

By performing the steps described above, the window parameter calculation operation is completed.

Next, the bather watching system 100 manages the state of the bather based on the classification. FIG. 15 is a flow chart showing an example of the operation for managing the bather's condition.

First, in step S210, the acquisition unit 11 acquires various data. In such a case, the acquisition unit 11 acquires environmental data of the bathroom 5 measured by the sensor 53, health data of the bather, and the like.

Next, in step S220, the state determination unit 17 determines the pre-acquired affiliation classification and bathing information based on one or more affiliation classifications classified in step S170 and bathing information combining environmental data and health data. The state data of the bather is determined by referring to a state database that records state data indicating the state of the bather for the information. Moreover, the state determination unit 17 may determine the state data based only on the affiliation classification without using the bathing information.

The status database is a database that records status data for belonging classification and bathing information. The state database records a state model generated using a plurality of pieces of learning data, with input data as belonging classification and bathing information, output data as state data, and a pair of input data and output data as learning data. You may In such a case, this state model differs from the above-described neural network model in that the input data are belonging classes and bathing information, and the output data are state data.

The state data is data indicating the state of the bather, such as "normal", "abnormal", "heatstroke", "myocardial infarction", and "fainting".

Further, in step S220, the state determination unit 17 may refer to the state database to determine the state data of the bather based on the plurality of affiliation categories classified in step S170. In such a case, the state determination unit 17 receives as input a plurality of affiliation classes estimated from a plurality of time-series continuous scalogram images, and determines state data. In such a case, for example, the state determination unit 17 may output "abnormal" state data when a certain number of premature ventricular contractions are observed among the classifications that are continuous in time series.

Next, in step S230, the treatment determination unit 16 determines treatment data for the state data. The treatment data is data indicating a treatment method for the state data, such as "drain the bathtub 51", "call an ambulance", and "notify the family".

The treatment determination unit 16 sets, for example, condition data obtained in advance and treatment data indicating a treatment method for the condition data as a pair of learning data for judgment, and creates a treatment information database in which a plurality of learning data for judgment are recorded. Refer to it to determine treatment data for the state data determined in step S220.

The treatment information database is a database that records treatment data for status data. The treatment information database uses input data as state data, output data as treatment data, a pair of input data and output data as learning data for judgment, and records a treatment model generated using a plurality of learning data for judgment. You may In such a case, this treatment model differs from the neural network model described above in that the input data is state data and the output data is treatment data.

Next, in step S240, the output unit 15 transmits the treatment data determined in step S230. In such a case, the output unit 15 transmits the treatment data to, for example, the terminal 2, the monitor 54, or the like, and presents the treatment data to the bather's family, caregiver, or bather. This makes it possible to present an appropriate treatment method to the bather's family, caregiver, or bather.

Next, in step S250, the drainage device 52 drains the bathtub 51 according to the treatment data transmitted in step S240. In such a case, the drainage device 52 may be set in advance as to whether or not to perform drainage for each treatment data. Further, the ventilation fan 56 may ventilate the inside of the bathroom 5 according to the treatment data. In such a case, whether or not to perform ventilation may be set in advance for each treatment data.

Next, in step S260, the acquisition unit 11 acquires result data indicating the treatment result for the treatment data determined in step S230. The result data is data indicating the result of treatment given to the bather, and includes data such as "no problem" or "problem". The acquisition unit 11 may acquire result data transmitted from the terminal 2, for example.

Next, in step S270, the storage unit 14 combines the state data determined in step S220 and the treatment data determined in step S230 into a pair of determination learning data according to the result data acquired in step S260. data and newly recorded in the treatment information database. For example, when the result data is "no problem", the storage unit 14 newly records the learning data for judgment in the treatment information database, and when the result data is "problem", the learning data for judgment is newly recorded in the treatment information database. It may not be recorded in the information database. Further, for example, if the result data is "no problem", the degree of association between the status data and the treatment data may be set to be high. As a result, it is possible to feed back the result, so that the accuracy can be further improved.

By performing the steps described above, the operation of the bather monitoring system 100 for managing the state of the bather is completed. This makes it possible to determine the state of the bather with high accuracy.

While embodiments of the invention have been described, the embodiments have been presented by way of example and are not intended to limit the scope of the invention. Such novel embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. This embodiment and its modifications are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and its equivalents.

Reference Signs List 1: heartbeat classification device 2: terminal 3: server 4: communication network 5: bathroom 10: housing 11: acquisition unit 12: conversion unit 13: affiliation classification estimation unit 14: storage unit 15: output unit 16: treatment determination unit 17 : State determination unit 18 : Extraction unit 19 : Separation unit 51 : Bathtub 52 : Drainage device 53 : Sensor 54 : Monitor 55 : Electrode 56 : Ventilation fan 100 : Bather monitoring system 101 : CPU
102: ROM
103: RAM
104: Storage unit 105: I/F
106: I/F
107: I/F
108: Input section 109: Display section 110: Internal bus

Claims (10)

In a heartbeat classification device that estimates a classification indicating a classification of each heartbeat of the electrocardiographic signal based on the electrocardiographic signal,
extracting means for extracting a partial signal containing a part of the electrocardiographic signal from the electrocardiographic signal;
probability calculating means for calculating the probability that the partial signal extracted by the extracting means is a reference partial signal containing the peak of the electrocardiographic signal;
Separating means for separating an electrocardiographic beat signal representing one cycle of the electrocardiographic signal based on the partial signal for which the probability calculated by the probability calculating means is equal to or greater than a threshold;
A heartbeat classification apparatus, comprising: classification affiliation estimation means for estimating the classification classification based on the electrocardiographic beat signal separated by the separation means. The probability calculation means sets the input data based on the reference partial signal and the probability associated with the input data as a pair of probability learning data, and the probability is generated by machine learning using a plurality of the probability learning data. 2. The heart rate classification device according to claim 1, wherein a probability is calculated based on the partial signals extracted by said extraction means with reference to a model. 2. The heartbeat classification device according to claim 1, wherein said extraction means extracts said partial signals based on heartbeat information about heartbeats. 2. The heart rate classification apparatus according to claim 1, wherein said extraction means newly extracts partial signals based on the electrocardiographic beat signal separated by said separation means. 2. The heart rate classification apparatus according to claim 1, wherein said separating means calculates the length of one cycle of said electrocardiographic beat signal based on distances between a plurality of peaks in said electrocardiographic signal. The classification estimating means converts the electrocardiogram beat signal separated by the separation means into a scalogram image, and combines input data based on the learning scalogram image acquired in advance and the classification associated with the input data into a pair. A neural network model generated using a plurality of the learning data as learning data is referred to, and the converted scalogram image is used as an input to estimate the belonging classification. 2. A heart rate classifier according to claim 1. a bathroom environment sensor that measures environmental data indicating any one or more of temperature, humidity, illuminance, reverberation, scent, and atmospheric pressure provided in the bathroom;
a bather information acquiring means for acquiring health data of a bather taking a bath in the bathroom;
Bathing information obtained by combining one or more affiliation classifications estimated by the heartbeat classification device according to claim 6, environmental data measured by the bathroom environment sensor, and health data obtained by the bather information obtaining means. a state determination means for determining the state data of the bather by referring to a state database in which state data indicating the state of the bather with respect to the previously obtained affiliation classification and bathing information is recorded based on;
State data obtained in advance and action data indicating a method of action for the state data are used as a pair of learning data for judgment, and referring to a action information database in which a plurality of learning data for judgment are recorded, the state judging means and treatment determination means for determining treatment data for the determined condition data. a transmitting means for transmitting treatment data determined by the treatment determining means to an external device;
8. The bather watching system according to claim 7, further comprising presenting means for presenting the treatment data determined by the treatment determining means to the bather. 8. The bather watching system according to claim 7, further comprising drain means for draining a bathtub provided in said bathroom based on the treatment data determined by said treatment determination means. a treatment result obtaining means for obtaining result data indicating a treatment result for the treatment data determined by the treatment determination means;
According to the result data acquired by the treatment result acquisition means, the state data determined by the state determination means and the treatment data determined by the treatment determination means are set as a pair of learning data for determination, and newly 8. The system for watching over a bather according to claim 7, further comprising recording means for recording in said treatment information database.

Images