JP2718302B2 - Sleep state determination device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sleep state determining means for determining a sleep state.

[0002]

2. Description of the Related Art The inventor of the present invention has devised a sleep state determining apparatus for detecting a sleep state of a sleeping person by detecting a body motion of the sleeping person. FIG. 12 shows a block diagram thereof. The output signal of the flexible piezoelectric element 1 disposed on the bedding is
The amplifying unit 3, the smoothing unit 4, and the comparing unit 5 perform processing to detect the body movement of the sleeping person, and the sleep judging unit 6 determines that the sleep state is established when the body motion is in a rest state for a certain period of time or longer. Is what you do. As a result, the sleep state can be easily and inexpensively detected, so that it is possible to apply the present invention to product fields such as environmental control of a bedroom and prevention of falling asleep.

[0003]

However, the sleep state detecting device of the above technology can determine that "the state of sleep has been reached (stages 2 to 4 in the sleep stage)".
There is a problem that determination cannot be made in terms of "at what level the current depth of sleep is". The reason for this is that, when judging the depth of sleep from body motion data, for example, a method of allocating the depth of sleep according to the frequency of occurrence of body motion per unit time can be considered. This is because the sleep state cannot be accurately determined by the above simple concept due to internal fluctuation.

Accordingly, a first object of the present invention is to estimate a sleep state by multidimensional information processing by means of a neural network model using body motion data during sleep.

A second and third object is to improve the accuracy of sleep state estimation by appropriately processing and selecting input information relating to body movement.

Fourth to twelfth objects are not only body movement data during bedtime but also other physical quantities that can be actually detected, that is, room temperature, room humidity, room air velocity, bed temperature, bed humidity, By using the illuminance, the noise level, the skin temperature, and the heart rate as input information for learning and input information for estimating a sleep state, the accuracy of sleep state estimation is further improved.

[0007]

In order to achieve the first object, the present invention provides a body movement detecting means for detecting a body movement of a human body, and a sleep state of the human body based on an output of the body movement detecting means. Sleep state estimating means, and the sleep state estimating means estimates a sleep state obtained by a method simulating a hierarchical neural network in which a plurality of layers composed of a plurality of neural elements are combined. And a neural network model having a plurality of connection weight coefficients of the neural network.

Further, in order to achieve the second object, the present invention provides a first operation for calculating an elapsed time from a point in time when a body movement occurred before a certain point in time based on an output of a body movement detecting means. Means and a storage means for storing the output of the first arithmetic means, and comprising a sleep state estimating means for estimating a sleep state using the output of the storage means as input information.

In order to achieve the third object, the present invention further comprises a second calculating means for calculating a frequency of occurrence of a body movement within a unit time based on an output of the body movement detecting means; And a sleep state estimating means for estimating a sleep state using the output of the storage means as input information.

In order to achieve the fourth object, the present invention has a room temperature sensor for detecting a room temperature, and a sleep state estimator for estimating a sleep state based on outputs of both a body movement detecting means and the room temperature sensor. Consisting of means.

In order to achieve the fifth object, the present invention has an indoor humidity sensor for detecting indoor humidity,
It comprises sleep state estimating means for estimating a sleep state based on outputs of both a body motion detecting means and the indoor humidity sensor.

According to another aspect of the present invention, there is provided an indoor airflow sensor for detecting an airflow velocity in a room, and a sleep state is detected based on outputs from both a body movement detecting means and the indoor airflow sensor. It comprises sleep state estimating means for estimating.

In order to achieve the seventh object, the present invention has a bed temperature sensor for detecting the temperature in the bed, and sleeps based on the outputs of both the body movement detecting means and the bed temperature sensor. It comprises sleep state estimation means for estimating the state.

In order to achieve the eighth object, the present invention has a bed humidity sensor for detecting the humidity in the bed, and sleeps based on the outputs of both the body movement detecting means and the bed humidity sensor. It comprises sleep state estimation means for estimating the state.

In order to achieve the ninth object, the present invention has an illuminance sensor for detecting illuminance in a room, and a sleep state for estimating a sleep state based on outputs of both a body movement detecting means and the illuminance sensor. It consists of state estimation means.

Further, in order to achieve the tenth object, the present invention has a noise sensor for detecting a noise level in a room, and estimates a sleep state based on outputs of both a body movement detecting means and the noise sensor. It comprises sleep state estimation means.

Further, in order to achieve the eleventh object, the present invention provides a skin temperature sensor for detecting a skin temperature of a sleeping person, and a sleep state estimated based on outputs of both a body movement detecting means and the skin temperature sensor. Sleep state estimating means.

Further, in order to achieve the twelfth object, the present invention provides a heart rate calculating means for calculating a heart rate based on the output of the body movement detecting means, It comprises sleep state estimating means for estimating a sleep state based on the output.

[0019]

According to the present invention, the sleep state estimating means estimates the sleep state of the sleeping person every moment by inputting the body movement data from the body movement detecting means to the sleep state estimating means. The neural network schematic means constituting the sleep state estimating means has a connection weight coefficient already learned in the sleep environment to be used, and estimates the sleep state of the sleeping person in the sleep environment every moment.

The body movement data from the body movement detecting means includes:
Various information is included in estimating the sleep state, and the body motion rest time and the occurrence frequency of the body motion are input to the sleep state estimating means.

Furthermore, not only the body movement data during bedtime but also other physical quantities that can be actually detected, that is, room temperature, indoor humidity, indoor air flow velocity, bed temperature, bed humidity, illuminance, noise level, skin temperature, The sleep state is accurately estimated based on the input information for learning the heart rate.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a perspective view when the present embodiment is mounted on a bed 26, and FIG. 2 is a block diagram. 1 and 2
In the above, 1 is a piezoelectric element, 2 is a filter, 3 is an amplifying means, 4 is a smoothing means, and 5 is a comparing means, which detects body movement with the above configuration. The piezoelectric element 1 is made of polyvinylidene fluoride (PV
DF) and a thin film made of a high-molecular piezoelectric material, and a flexible electrode film is adhered to both sides to form a tape.
Is fixed to the surface of the mattress 27 as shown in FIG. The comparing means 5 determines three states of absence, presence, and movement of the human body in accordance with the output signal of the smoothing means 4, and outputs three output ports (absence output port 5a, presence output port 5b). And a body motion output port 5c). Reference numeral 7 denotes sleep state estimating means for estimating a sleep state based on a signal from the comparing means 5, and a neural network schematic means 8, a signal processing means 9,
It has storage means 10. Reference numeral 11 denotes a timing unit that starts timing based on the output of the comparison unit 5, and 12 denotes a display unit that displays a result estimated by the sleep state estimation unit 7. The filter 2, the amplifying unit 3, the smoothing unit 4, the comparing unit 5, the sleep state estimating unit 7, and the timing unit 11 are built in the unit 13. The piezoelectric element 1 and the unit 13 are connected by a shield wire 14.

According to the configuration of this embodiment, the present invention operates as follows. When a human body is present on the bedding and the piezoelectric element 1 is deformed by the body movement of the human body, piezoelectric is generated from the piezoelectric element according to the degree of the deformation. This output signal is filtered by the filter 2, and the signal is amplified by the amplifying means 3 and smoothed by the smoothing means 4.
The output waveform from the smoothing means 4 is shown in FIG. In the figure, the S portion is an output due to a small body motion (hereinafter referred to as fine body motion) caused by the heartbeat / respiratory activity in a resting state, and the G2 portion is due to a relatively large motion (hereinafter referred to as coarse body motion) such as turning over. Represents the output. In accordance with the output signal level V of the smoothing means 4, the comparing means 5 operates as follows. 1) V
Is less than a predetermined first set value Va, it is determined that there is no person on the bedding, and the output level of the absence determination output port 5a is set to Hi. 2) If V is equal to or greater than Va and less than a second predetermined value Vb, it is determined that a person exists on the bedding in a resting state, and the output level of the occupancy determination output port 5b is set to Hi. 3) If V is equal to or higher than Vb, it is determined that the human body has moved (coarse body movement) on the bedding, and the output level of the body movement determination output port 5c is set to Hi.

Next, the sleep state estimating means 7 estimates the sleep state based on the outputs of the three output ports. Here, regarding the behavior of body movements during sleep, it is well known empirically and experimentally that, when the state of falling asleep is reached, the body movement disappears, and when the sleep state becomes shallow, there are many body movements, for example, The number of occurrences of body movements per unit time may be calculated, and the sleep state may be estimated based on the calculation. Due to internal fluctuations, the sleep state cannot be accurately determined by the above simple concept. Therefore, the means constituting the sleep state estimating means 7 is constituted by a method simulating a neural network which is optimal as a multidimensional information processing technique. In a method simulating a neural network, there are a method of using a plurality of connection weight coefficients of the neural network for estimating a sleep state as a fixed table, and a method of leaving a learning function and adapting to an environment and a user. is there. In the present embodiment, the sleep state estimating means 7 having a neural network model having a fixed connection weight coefficient therein for estimating a sleep state obtained by a method imitating a neural network is provided.

The fixed connection weight coefficient in the neural network for estimating the sleep state is obtained, for example, by performing a polygraph measurement during sleep to collect data on sleep stages and body movements,
The correlation between the sleep stage data and the body motion data can be obtained by learning the neural network model means. For the neural network model to be used, reference 1 (PDP model, DE Ramelhart et al., Translated by Shunichi Amari, 2
989), Document 2 (Basics of neurocomputer, P
102, Kaoru Nakano and 7 others, 1990), Japanese Examined Patent Publication 63-5
There is a technique disclosed in Japanese Patent No. 5106 and the like. Hereinafter, the configuration and operation of specific neural network model means will be described using a multilayer perceptron using an error back propagation method as an example of the most well-known learning algorithm described in Document 1. FIG. 4 is a conceptual diagram of a neural element which is a constituent unit of the neural network schematic unit 8. In FIG.
8n are pseudo-synaptic connection converters that simulate nerve synaptic connections, and 8a is a pseudo-synaptic connection converter 8l-
8b is an adder that adds the outputs from 8n, 8b is a set nonlinear function, for example, a sigmoid function with h as an axle value, f (y, h) = 1 / (l + exp (−y + h)) (Equation 1) Is a non-linear converter for non-linearly converting the output of the adder 8a. The input lines for receiving the correction signal from the correction means are connected to the pseudo-synaptic coupling converters 8l to 8n and the non-linear converter 8b, although they are omitted because the drawing becomes complicated. The pseudo synapse connection converters 8l to 8n serve as connection weight coefficients of the neural network model means 8. This neural element has two types of operation modes, a signal processing mode and a learning mode. Hereinafter, the operation of each mode of the neural element will be described with reference to FIG. First, the operation in the signal processing mode will be described. The neural element has n inputs Xl to Xn
And outputs one output. The i-th input signal Xi is
It is converted to Wi · Xi in the i-th pseudo-synaptic connection converter 8i shown by a square. N signals Wl.Xl to Wn converted by the pseudo-synaptic coupling converters 8l to 8n
Xn enters the adder 8a shown by a circle, the addition result y is sent to the nonlinear converter 8b shown by an ellipse, and the final output f
(Y, h). Next, the operation in the learning mode will be described. In the learning mode, the pseudo-synaptic connection converter 8l
8n and the conversion parameters Wl to Wn of the nonlinear converter 8b
And h, the conversion parameter correction amount ΔWl from the correction means
Receiving a correction signal representing .DELTA.Wn and .DELTA.h, Wi + .DELTA.Wi; i = 1, 2,..., Nh + .DELTA.h (equation 2). FIG. 5 is a conceptual diagram of a signal conversion unit in which four of the above neural elements are connected in parallel. Needless to say, the following description does not specify the number of neural elements constituting the signal conversion means as four. In FIG. 5, reference numerals 811 to 844 denote pseudo-synaptic coupling converters,
Reference numerals 801 to 804 denote add-nonlinear converters that combine the adder 8a and the non-linear converter 8b described with reference to FIG. In FIG. 5, an input line for receiving a correction signal from the correction means is connected to the pseudo-synaptic coupling converters 811 to 844 and the addition non-linear converters 801 to 804, as in FIG. The pseudo synapse connection converters 811 to 844 also become connection weight coefficients. Regarding the operation of this signal conversion means, the operation of the neural element described in FIG. 4 is performed in parallel. FIG.
FIG. 4 is a block diagram showing a configuration of a signal processing unit 9 when an error back propagation method is adopted as a learning algorithm. FIG.
In the figure, reference numeral 91 denotes the signal conversion means described above, however, here, m neural elements receiving n inputs are arranged in parallel, and 92 calculates the correction amount of the signal conversion means 91 in the learning mode. Correction means. Hereinafter, FIG.
The operation in the case where the learning of the signal processing means is performed based on the above will be described. The signal conversion means 91 has n input Sin
(X), and outputs m output Sout (X).
The correction means 92 includes an input signal Sin (X) and an output signal So.
ut (X), and waits for the input of m error signals δi (X) from the error calculation means or the signal conversion means at the subsequent stage. The error signal δi (X) is input and the correction amount is set as ΔWij = δi (X) · Siout (X) · (1-Siout (X)) · Sjin (X) (i = 1 to n, j = 1 to m) Then, the correction signal is sent to the signal conversion means 91. The signal conversion means 91 corrects the conversion parameters of the internal nerve elements according to the learning mode described above. FIG. 7 is a block diagram showing a configuration of a multilayer perceptron using a neural network model. In FIG. 7, 91X, 91
Y and 91Z are signal conversion means composed of k, l and m neural elements, respectively, 92X, 92Y and 92Z are correction means, and 933 is an error calculation means. The operation of the multilayer perceptron configured as described above will be described with reference to FIG. In the signal processing means 94X, the signal conversion means 91X includes an input Siin (X)
(I = 1 to n), and outputs Sjout (X) (j = 1
To k). The correcting means 92X outputs the signal Siin
(X) and the signal Sjout (X), and an error signal δj
(X) Wait until (j = 1 to k) is input. Hereinafter, the same processing is performed in the signal processing means 94Y and 94Z, and the final output Shot is outputted from the signal conversion means 91Z.
(Z) (h = 1 to m) is output. Final output Show
t (Z) is also sent to the error calculating means 93. In the error calculating means 93, an ideal output T (T1,..., T
M) is calculated, and an error signal δh (Z) is sent to the correcting means 92Z.

[0026]

(Equation 1)

Η is a square error as a parameter evaluation function that determines the learning speed of the multilayer perceptron, and the error signal is δh (Z) = − η · (Shout (Z) −Th) (Equation 5) . The correction unit 92Z calculates the correction amount ΔW (Z) of the conversion parameter of the signal conversion unit 91Z according to the procedure described above, calculates the error signal to be sent to the correction unit 92Y based on (Equation 6), and calculates the correction signal. ΔW (Z) is sent to the signal conversion means 91Z, and the error signal δ (Y) is corrected by the correction means 92Z.
Send to Y The signal conversion means 91Z outputs the correction signal ΔW (Z)
Modify internal parameters based on. Note that the error signal δ (Y) is given by (Equation 6).

[0028]

(Equation 2)

Here, Wij (Z) is the signal conversion means 91
Z is a conversion parameter of the pseudo-synaptic coupling converter of Z.

Hereinafter, similar processing is performed by the signal processing means 91X,
This is performed at 91Y. By repeatedly performing the above procedure called learning, the multilayer perceptron, when given an input, outputs an output that approximates the ideal output T well. In the above description, a three-stage multilayer perceptron is used, but the number of stages may be any.
Also, the method of correcting the conversion parameter h of the non-linear conversion means in the signal conversion means and the method of speeding up the learning known as the inertia term in the document conversion means are omitted for simplification of description, but this omission is omitted. The present invention described below is not restricted.

Thus, the neural network model 8 learns the relationship between the sleep stage data and the body motion data, and can express in a natural way an estimation method that cannot be easily described by simple rules. In the present embodiment, the sleep state estimating means 7 is configured by incorporating the information thus obtained. More specifically, the sleep state estimating means 7 is constituted by using only the signal converting means 91X, 91Y, and 91Z of the multilayer perceptron after sufficiently learning as the neural network schematic means 8.

The data actually learned will be described. FIG. 8 shows overnight sleep stage data of a subject A in an overnight sleep experiment. FIG. 9 shows sleep stage data of another sleeping person of the same subject A. FIG. 10 shows sleep stage data of another subject B. In FIGS. 8 to 10, the portion marked as the output of the body motion corresponds to the case where the output of the body motion output port 5c of the comparison means 5 described above is Hi. As shown in FIG. 8 to FIG. 10, variation between individuals and variation between individuals are recognized between data. This is due to the fact that the experimental results vary depending on the physical and psychological state of the subject and the sleeping environment conditions. The purpose of this embodiment is to make such inter-individual variation and individual variation variation predictable by learning. For this reason, the above-mentioned all-night sleep experiment was performed several times for each of a plurality of subjects, and the experimental data obtained from each of the subjects was input to the neural network model means for learning. However, in this experiment, environmental conditions (temperature and humidity, illuminance, noise, bedding, etc.) were the same in all trials. The neural network schematic means 8 receives and learns two types of information, that is, time series data of body movements, data of elapsed time from the time of entering the bed, and sleep stage data of the subject as an ideal output, because of the ease of input. The signal conversion means 91X, 91Y, 91Z in the neural network schematic means 8 were established, and they were incorporated into the sleep state estimating means 7 as the neural network schematic means 8. Here, the time series data of the body movement is the body movement output port 5 of the comparing means 5 described further.
This is time-series data of a binary signal such as 1 if the output of c is Hi and 0 if Lo.

Next, the operation of this embodiment will be described with reference to the block diagram shown in FIG. When the power is turned on, the absence, occupancy, and body movement are determined by the comparing means 5 based on the operation described above, and the output state of each output port of absence, occupancy, and body movement is displayed on the display means 12. When the sleeping person enters the bed, the presence of the bed is determined by the comparing means 5 and the bed output port becomes Hi. At this point, the timing operation of the timing means 11 is started. That is, the timer 11 measures the elapsed time from entering the bed. The output of the body movement output port 5c of the comparison means 5 is stored in the storage means 10. Outputs from the storage unit 10 and the timer unit 11 are input to the sleep state estimation unit 7. The sleep state estimating means 7 estimates the sleeping state of the sleeping person from time to time based on the input information, and outputs the information to the display means 12. The above operation is stopped when the sleeping person leaves the bed and the output of the absence output port 5a becomes Hi. If the user leaves the bed during bedtime due to getting up in the toilet or the like, the above operation is restarted by entering the bed again. To clear the display on the display means 12, the power may be turned on again.

As described above, according to the present embodiment, there is provided a sleep state estimating unit incorporating a neural network model having a plurality of fixed connection weight coefficients of a neural network which has already been learned in an environment in which it is used. With this configuration, the sleep state can be accurately determined even if there is inter-individual variation or intra-individual variation.

Next, a second embodiment of the present invention will be described. This embodiment is different from the first embodiment in that the sleep state estimating means 7 is based on the output of the body movement detecting means 6 at a certain time Tx and at a time Txo before the occurrence of the body movement.
A first calculating means 15 for calculating the elapsed time t from the storage means 10 and a storage means 10 for storing the elapsed time t.
The sleep state is estimated using the output of 0 as input information. It can be seen from a sleep experiment or the like that the longer the time during which no body motion occurs (hereinafter referred to as the body motion quiescent time), the deeper the sleep state at that time. That is, the present embodiment aims to improve the accuracy of sleep state determination by inputting not only information such as the presence or absence of body movement but also the body movement stationary time.

The neural network schematic means 8 includes time series data of body motion, data of elapsed time from the time of entering the bed, and time of elapsed time t from the time Txo when the body motion occurred before a certain time Tx. The three information of the series data and the sleep stage data of the subject as an ideal output are input and learned, and the signal conversion means 91X, 91Y, 9 in the neural network schematic means 8 are input.
1Z were established and incorporated into the sleep state estimating means 7 as a neural network schematic means 8. Then, the same operation as that of the first embodiment is performed, and the peripheral configuration can be achieved by the configuration of the first embodiment. This further improves the sleep state estimation accuracy.

Next, a third embodiment of the present invention will be described.
This embodiment is different from the above-described embodiment in that the sleep state estimating unit 7 calculates the frequency of occurrence of body movement within a unit time based on the output of the body movement detection unit 6, And a storage means 10 for storing the output of the arithmetic means 16 of the present invention, and the sleep state is estimated using the output of the storage means 10 as input information. Sleep experiments and the like show that if the frequency of occurrence of body motion is high, the sleep state is often light, whereas if the frequency of occurrence of body motion is low, the sleep state is often high.
That is, the present embodiment aims to improve the accuracy of sleep state determination by inputting not only information such as the presence or absence of body movement but also the frequency of occurrence of body movement.

The neural network schematic means 8 includes three pieces of information, namely, time series data of body movements, elapsed time data from the time of entering the bed, and time series data on the frequency of occurrence of body movements per unit time.
The sleep stage data of the subject is input and learned as an ideal output, and the signal conversion unit 91X in the neural network schematic unit 8 is used.
91Y and 91Z were established, and they were incorporated into the sleep state estimating means 7 as the neural network schematic means 8. And the first
The same operation as that of the first embodiment is performed, and the peripheral configuration can be achieved by the configuration of the first embodiment. This further improves the sleep state estimation accuracy.

Further, at the time of learning in the above embodiment, a total of four pieces of information to which the information of the frequency of occurrence of body motion T time before the present time is added as input information, and sleep stage data of the subject as ideal output are input. Learning, signal conversion means 91X, 91Y, 91 in the neural network schematic means 8
Z may be established and incorporated into the sleep state estimating means 7 as the neural network schematic means 8, further improving the sleep state estimation accuracy.

Next, a fourth embodiment of the present invention will be described.
This embodiment is different from the above-described embodiment in that the sleep state estimating unit 7 estimates the sleep state using the output of the room temperature sensor 17 for detecting the room temperature and the output of the body movement detecting unit 6 as input information. Here, the room temperature sensor 17 is installed in the unit 12 as shown in FIG. 11, but may be installed on the indoor wall surface or the like. When the room temperature exceeds about 25 ° C. by a sleep experiment or the like, the higher the room temperature, the more frequently the body motion occurs and the more often a light sleep state appears. That is, the present embodiment aims to improve the accuracy of sleep state determination by inputting not only information such as presence / absence of body movement but also information on room temperature.

With the above configuration, the sleep experiment was performed several times for a plurality of subjects each night, and the experimental data obtained from each subject was input to the neural network model means 8 for learning.
However, in this experiment, environmental conditions other than room temperature (humidity, illuminance,
The experiment was performed under a plurality of room temperature conditions under the same conditions for all trials. The neural network schematic means 8 receives the three information of the time series data of the body movement obtained from the above experiment, the elapsed time data from the time of entering the bed, the room temperature data from the room temperature sensor 17, and the sleep of the subject as an ideal output. The stage data was input and learned, and the signal conversion means 91X, 91Y, and 91Z in the neural network schematic means 8 were established, and they were incorporated in the sleep state estimating means 7 as the neural network schematic means 8.

As a result, when the sleeping person enters the bed, the presence of the bed is determined by the comparing means 5 based on the operation described above, and the output port in the bed becomes Hi, at which time the timing operation of the timing means 11 is started. . The output of the body movement output port 5c of the comparison means 5 is stored in the storage means 10, and the outputs from the storage means 10, the timing means 11, and the room temperature sensor 17 are input to the sleep state estimation means 7. The sleep state estimating means 7 estimates the sleeping state of the sleeping person from time to time based on the input information, and outputs the information to the display means 12.
As described above, in the present embodiment, the accuracy of the sleep state determination can be improved by inputting not only the information such as the presence or absence of body movement but also the information of the room temperature.

In the fourth embodiment, the sleep state is estimated by detecting the room temperature information in addition to the body motion information.
As a fifth embodiment of the present invention, an indoor humidity sensor 18 may be provided as shown in FIG. 11 to detect indoor humidity information in addition to body motion information to estimate a sleep state. Learning procedure Same as the fourth embodiment, but in experiments for obtaining data, environmental conditions (humidity, illuminance, noise,
Bedding, etc.) were subjected to experiments under a plurality of indoor humidity conditions under the same conditions in all trials, and the neural network schematic means 8 received time series data of body movement obtained from the above experiments, The time data, the room temperature data from the indoor humidity sensor 18 and the sleep stage data of the subject as an ideal output are input and learned, and the signal conversion means 91X, 91Y, and 91Z in the neural network schematic means 8 are established. These are incorporated into the sleep state estimating means 7 by means of a neural network schematic means 8.

FIG. 1 shows a sixth embodiment of the present invention.
As shown in FIG. 1, the indoor airflow sensor 19 may be provided to detect the airflow velocity information of the sleeping place in addition to the body motion information to estimate the sleep state. The learning procedure is the same as in the above embodiment.

FIG. 1 shows a seventh embodiment of the present invention.
As shown in FIG. 1, a configuration may be adopted in which the in-bed temperature sensor 20 is provided to detect the in-bed temperature information in addition to the body movement information to estimate the sleep state. The learning procedure is the same as in the above embodiment.

As an eighth embodiment of the present invention, FIG.
As shown in FIG. 1, a configuration may be adopted in which the in-bed humidity sensor 21 is provided to detect the in-bed temperature information in addition to the body movement information to estimate the sleep state. The learning procedure is the same as in the above embodiment.

As a ninth embodiment of the present invention, FIG.
As shown in FIG. 1, an illuminance sensor 22 may be provided to detect the illuminance information in addition to the body motion information to estimate the sleep state. The learning procedure is the same as in the above embodiment.

As a tenth embodiment of the present invention, as shown in FIG. 11, a noise sensor 23 may be provided to detect the noise level information in addition to the body motion information to estimate the sleep state. The learning procedure is the same as in the above embodiment.

Further, as an eleventh embodiment of the present invention, as shown in FIG. 11, a skin temperature sensor 24 may be provided to detect the skin temperature information in addition to the body motion information to estimate the sleep state. . The learning procedure is the same as in the above embodiment.

Further, as a twelfth embodiment of the present invention, based on the knowledge that the heart rate decreases and stabilizes as the sleep becomes deeper, The heart rate calculating means 25 for calculating the heart rate may be provided in the unit 11, and the sleep state may be estimated by detecting the heart rate information in addition to the body motion information. The learning procedure is the same as in the above embodiment. The calculation of the heart rate can be obtained, for example, by calculating the autocorrelation of the output of the smoothing means 4 and obtaining its peak value.

In the above embodiment, two or three pieces of information including body motion information are input. However, the present invention is not restricted to the present invention, and is used for example in the fourth to twelfth embodiments. Physical quantities: room temperature, indoor humidity, indoor airflow velocity,
Bed temperature, bed humidity, illuminance, noise level, skin temperature,
A configuration may be adopted in which a plurality of heart rates are detected, and input together with body motion information. Thereby, the accuracy of sleep state estimation can be further improved.

[0052]

As described above, according to the structure of the present invention, the following effects can be obtained. (1) Since the system is provided with the sleep state estimating means incorporating the neural network model having a plurality of fixed connection weighting factors of the neural network which has already been learned in the environment in which it is used, variation between individuals and individual The sleep state can be determined with high accuracy even if there is internal fluctuation. (2) Since it has a configuration including sleep state estimating means incorporating neural network model means obtained by learning using not only information such as the presence or absence of body motion but also body motion rest time as input information, the sleep state The accuracy of the determination is further improved. (3) Since it has a configuration including sleep state estimating means incorporating neural network model means obtained by learning using not only information such as presence / absence of body movement but also occurrence frequency of body movement as input information, The accuracy of the state determination is further improved. (4) In addition to the body motion data during sleep, other physical quantities that can be actually detected, that is, room temperature, indoor humidity, indoor air flow velocity,
Bed temperature, bed humidity, illuminance, noise level, skin temperature,
Since the heart rate is input, the accuracy of sleep state estimation is further improved.

[Brief description of the drawings]

FIG. 1 is a perspective view of a sleep state determination device according to a first embodiment of the present invention.

FIG. 2 is a block diagram of the apparatus.

FIG. 3 is an output waveform diagram of a smoothing means in the apparatus.

FIG. 4 is a conceptual diagram of a neural element which is a constituent unit of a neural network schematic unit of the apparatus.

FIG. 5 is a conceptual diagram of a signal conversion unit configured by connecting four neural elements of the device in parallel.

FIG. 6 is a block diagram showing a configuration of a signal processing unit employing an error back propagation method as a learning algorithm of the apparatus.

FIG. 7 is a block diagram showing the configuration of a multilayer perceptron using the neural network model of the apparatus.

FIG. 8: Sleep stage data of a subject A in an overnight sleep experiment

FIG. 9 shows sleep stage data of a certain subject A at another bedtime in an overnight sleep experiment.

FIG. 10 shows sleep stage data of a subject B in an overnight sleep experiment.

FIG. 11 is a perspective view of a sleep state determination device according to another embodiment of the present invention.

FIG. 12 is a block diagram of a conventional sleep state determination device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Piezoelectric element 2 Filter 3 Amplification means 4 Smoothing means 5 Comparison means 6 Sleeping judgment means 7 Sleep state estimation means 8 Neural network model means 9 Signal processing means 10 Storage means 11 Clocking means 12 Display means 17 Room temperature sensor 18 Indoor humidity sensor 19 indoor airflow sensor 20 temperature sensor in bed 21 humidity sensor in bed 22 illuminance sensor 23 noise sensor 24 skin temperature sensor

Claims (12)

(57) [Claims] A sleep state estimating means for estimating a sleep state based on an output of the body movement detecting means, wherein the sleep state estimating means comprises a plurality of nerves; A neural network model having a plurality of connection weighting factors of a fixed neural network for estimating a sleep state obtained by a method simulating a hierarchical neural network in which many layers composed of elements are combined A sleep state determination device having: 2. A first calculating means for calculating an elapsed time from a point in time at which a body movement occurs before a certain time based on an output of the body movement detecting means, and an output of the first calculating means is stored. The sleep state determination means according to claim 1, further comprising a storage means, and comprising sleep state estimation means for estimating a sleep state using an output of said storage means as input information. A second calculating means for calculating a frequency of occurrence of a body movement within a unit time based on an output of the body movement detecting means; a storage means for storing an output of the calculating means; The sleep state determining means according to claim 1, comprising sleep state estimating means for estimating a sleep state by using the output of (1) as input information. 4. The sleep state determining means according to claim 1, further comprising a room temperature sensor for detecting a room temperature, and comprising sleep state estimating means for estimating a sleep state based on outputs of both the body movement detecting means and the room temperature sensor. 5. The sleep state according to claim 1, further comprising an indoor humidity sensor for detecting indoor humidity, and sleep state estimating means for estimating a sleep state based on outputs of both the body movement detecting means and the indoor humidity sensor. Judgment means. 6. The sleep according to claim 1, further comprising an indoor airflow sensor for detecting an airflow velocity in the room, and sleep state estimating means for estimating a sleep state based on outputs of both the body movement detecting means and the indoor airflow sensor. State determination means. 7. A sleep state estimating means having an in-bed temperature sensor for detecting a temperature in the bed, and a sleep state estimating means for estimating a sleep state based on outputs of both the body movement detecting means and the in-bed temperature sensor. Sleep state determining means. 8. A sleep state estimating means having a bed humidity sensor for detecting the humidity in the bed and estimating a sleep state based on outputs of both the body movement detecting means and the humidity sensor in the bed. Sleep state determining means. 9. The sleep state determining means according to claim 1, further comprising an illuminance sensor for detecting illuminance in the room, and a sleep state estimating means for estimating a sleep state based on outputs of both the body motion detecting means and the illuminance sensor. . 10. The sleep state determination according to claim 1, further comprising a noise sensor for detecting a noise level in the room, and sleep state estimation means for estimating a sleep state based on outputs of both the body motion detection means and the noise sensor. means. 11. The sleep state according to claim 1, comprising a skin temperature sensor for detecting a skin temperature of a sleeping person, and sleep state estimation means for estimating a sleep state based on outputs of both a body movement detection means and the skin temperature sensor. Judgment means. 12. Heart rate calculating means for calculating a heart rate based on the output of the body movement detecting means, and sleep state estimating means for estimating a sleep state based on the outputs of both the body movement detecting means and the heart rate calculating means. The sleep state determining means according to claim 1.