JP2021048961A - Sleep state estimation apparatus, sleep state estimation method and program - Google Patents

Abstract

To make it possible to appropriately standardize a feature amount or the like to thereby accurately estimate a sleep state of a subject on the basis of the standardized feature amount or the like.SOLUTION: A sleep state estimation apparatus 100 comprises: an acquisition part 11 for acquiring a biological signal parameter based on a biological signal of a subject; a determination part 12 for determining whether a certain sleep state appears as a sleep state of the subject or not; a generation part 13 for generating a parameter for standardization on the basis of biological signal parameters acquired until when the determination part 12 determines that the certain sleep state appears as the sleep state of the subject; a standardization part 14 for standardizing a biological signal parameter on the basis of the generated parameter for standardization; and an estimation part 15 for estimating the sleep state of the subject on the basis of the standardized biological signal parameter.SELECTED DRAWING: Figure 1

Description

The present invention relates to a sleep state estimation device, a sleep state estimation method and a program.

Conventionally, as a device for estimating a person's sleep state, for example, a device disclosed in Patent Document 1 is known. This conventional device includes a sensor sheet laid on a bed containing a plurality of piezoelectric elements, and the time required for the human breathing signal detected by this sensor to change from the maximum to the minimum is calculated as the expiratory time. Then, the average value of the expiratory time per 5 minutes is standardized, and the sleep state of the person is estimated based on the standardized average value. Also, with this device, when it is initialized or for individual users, if you try to standardize from the beginning, you can not estimate the sleep state in real time, so at first compare the general value with the average value It is desirable to estimate the sleep state and shift to the standardized value judgment after several hours have passed.

Japanese Unexamined Patent Publication No. 2008-68019

In this way, in the conventional device, when the initial setting is performed, the sleep state is estimated by comparing the general value with the average value, and the standardized value is used when several hours have passed from the start of sleep. It is said that it is desirable to shift to judgment, but there was no disclosure about when to shift to judgment based on standardized values. If the timing of shifting to the judgment based on the standardized value is not correct, the standardization cannot be performed properly, and as a result, the sleep state of the person may not be estimated accurately.

The present invention has been made in view of the above circumstances, and it is possible to appropriately standardize the feature amount and the like, thereby accurately estimating the sleep state of the subject based on the standardized feature amount and the like. It is an object of the present invention to provide a sleep state estimation device, a sleep state estimation method, and a program capable of performing the above.

In order to achieve the above object, the sleep state estimation device according to the present invention
An acquisition means for acquiring biological signal parameters based on the target biological signal, and
A determination means for determining whether or not a certain sleep state has appeared as the target sleep state, and
A generation means for generating standardization parameters based on the biological signal parameters up to the time when it is determined by the determination means that a certain sleep state has appeared as the target sleep state.
A standardization means for standardizing the biological signal parameter based on the generated standardization parameter,
An estimation means for estimating the sleep state of the subject based on the standardized biological signal parameters, and
To be equipped.

According to the present invention, it is possible to appropriately standardize the feature amount and the like, thereby accurately estimating the sleep state of the target based on the standardized feature amount and the like.

It is a figure which shows the structural example of the sleep state estimation apparatus which concerns on 1st Embodiment. It is a figure which shows an example of a pulse wave. It is a figure which shows an example of the estimated heartbeat interval. It is a figure which shows an example of resampling of a heartbeat interval. It is a figure explaining the time window used for the extraction of a data string. It is a flowchart of the sleep state estimation process which concerns on 1st Embodiment. It is a flowchart of the sleep state estimation process which concerns on modification 1. It is a flowchart of the sleep state estimation process which concerns on modification 2.

Hereinafter, embodiments will be described with reference to the drawings. The same or corresponding parts in the figure are designated by the same reference numerals.

(First Embodiment)
The sleep state estimation device 100 according to the first embodiment is a device that acquires a biological signal of a human subject as a target by a sensor and estimates the sleep state of the subject using the acquired biological signal. As shown in FIG. 1, the sleep state estimation device 100 includes a control unit 10, a storage unit 20, a sensor unit 30, an input unit 41, an output unit 42, and a communication unit 43 as functional configurations. ..

The control unit 10 is composed of a CPU (Central Processing Unit) or the like, and by executing a program stored in the storage unit 20, each unit (acquisition unit 11, determination unit 12, generation unit 13, standardization unit 14, etc.) described later will be executed. The function of the estimation unit 15) is realized. The control unit 10 also has a function of measuring time.

The storage unit 20 is composed of a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The ROM stores a program executed by the CPU of the control unit 10 and data necessary for executing the program in advance. Data that is created or changed during program execution is stored in the RAM.

The sensor unit 30 includes a pulse wave sensor that detects a volume pulse wave and an acceleration sensor that detects body movement, and these sensors are attached to, for example, the ears of a subject. The pulse wave sensor irradiates light from the surface of the subject's skin and observes the reflected light and transmitted light to detect a change in the amount of light absorbed by hemoglobin oxide in the blood. Since this change in the amount of light absorbed corresponds to the change in the volume of the blood vessel, the pulse wave sensor can obtain the pulse wave 201 in which the change in the volume of the blood vessel is captured as a waveform, as shown in FIG.

Since it is considered that the change in blood vessel volume is synchronized with the heartbeat, the timing at which the pulse wave peaks is defined as the timing when the heartbeat occurs (hereinafter referred to as "beating timing"). It can be estimated, and the time interval between two peaks of pulse waves adjacent to each other on the time axis can be estimated as a heartbeat interval (RR Interval: RRI). The pulsation timing can be estimated by using the timer (or clock) of the CPU or the sampling frequency of the pulse wave, but it may be estimated by another appropriate method. Therefore, the control unit 10 can estimate the heartbeat interval based on the pulse wave detected by the sensor unit 30. In FIG. 2, the timing at which the pulse wave 201 peaks corresponds to the pulsation timings 201t, 202t, 203t, and the peak interval of the pulse wave 201 corresponds to the heartbeat intervals 202i, 203i.

In addition, the acceleration sensor detects the body movement of the subject by detecting the acceleration in each of the three axes (X-axis, Y-axis, and Z-axis orthogonal to each other).

The input unit 41 is composed of, for example, a keyboard, a mouse, a touch panel, and the like. The input unit 41 is an interface for receiving a user operation such as an instruction to start / end the sleep state estimation process, for example.

The output unit 42 includes, for example, a display unit such as an LCD (Liquid Crystal Display) or an EL (Electroluminescence) display, a sound output unit such as a speaker or a buzzer, and a vibrating unit such as a vibrator. However, the output unit 42 may include only a part of these. The output unit 42 displays, for example, the sleep state estimated by the sleep state estimation device 100. Further, when the sleep state estimation device 100 is used as an alarm clock, the output unit 42 makes a sound at a designated time, for example.

The communication unit 43 is a communication interface for exchanging data and the like with other external devices. This communication interface may be wireless or wired. For example, the sleep state estimation device 100 can transmit the estimated sleep state to an external server or the like via the communication unit 43.

Next, the functional configuration of the control unit 10 of the sleep state estimation device 100 will be described. The control unit 10 realizes the functions of the acquisition unit 11, the determination unit 12, the generation unit 13, the standardization unit 14, and the estimation unit 15, and estimates the sleep state of the subject. In the present embodiment, the sleep states are wake (wake state), REM (REM sleep state), N1 (light sleep state among non-REM sleep states), and N2 (medium depth sleep among non-REM sleep states). State) and N3 (a state of deep sleep among non-REM sleep states) are treated as if they exist. However, regarding the sleep state, it is only necessary to distinguish between the REM sleep state and the non-REM sleep state, and it is arbitrary how many states the non-REM sleep state is distinguished from.

The acquisition unit 11 is a feature amount (RRI system feature amount, body movement system feature amount, respiratory system feature amount) calculated from biological signals (pulse wave, body movement (acceleration), etc.) detected by the sensor included in the sensor unit 30. ) Is obtained. Here, the RRI system feature amount is a feature amount related to the heartbeat interval of the subject, the body movement system feature amount is a feature amount related to the body movement of the subject, and the respiratory system feature amount is a feature amount related to the subject's breathing. It is a feature quantity. The acquisition unit 11 functions as an acquisition means.

The operation of the acquisition unit 11 will be described in more detail. The acquisition unit 11 acquires the pulse wave data string by sampling the detection value of the pulse wave sensor among the detection values obtained by the sensor unit 30 at a certain sampling frequency (for example, 250 Hz). Then, by detecting the peak of the amplitude from the pulse wave data sequence, the beat timing is estimated in time series, and the time interval between two beat timings adjacent to each other on the time axis is set as the heartbeat interval (RRI). Estimate as. When the estimated heartbeat intervals are arranged with time on the horizontal axis and heartbeat intervals on the vertical axis (interval between the beat timing on the horizontal axis and the beat timing immediately before it), a graph as shown in FIG. 3 is obtained. Be done.

FIG. 3 shows an example in which the original pulsation between the pulsation timing 204t and the pulsation timing 205t could not be estimated. In this case, the heartbeat interval 205i is regarded as an outlier (outlier value). Will be removed. Any method can be used to remove the outliers, but it is easy to assume that a heartbeat interval that does not fall within a certain time range, for example, a heartbeat interval of less than 0.5 seconds or 2 seconds or more, is an impossible value. There is a method of removing it.

Then, the acquisition unit 11 resamples the heartbeat interval data string after removing the outliers at a resampling frequency (for example, 2 Hz) lower than the above sampling frequency to generate an evenly spaced RRI data string. , Get the generated data string. The equidistant RRI is a data sequence obtained by interpolating a data string of heartbeat intervals and resampling the data at equal intervals in time. Since the heartbeat fluctuates and the heartbeat interval data string is not acquired at equal intervals in time, it is difficult to use it as it is. Therefore, in the present embodiment, the data string of the equidistant RRI is acquired from the data string of the heartbeat interval, and the feature amount is calculated using the data string of the equidistant RRI. Any method can be used to acquire the data string of the evenly spaced RRI. For example, as shown in FIG. 4, the dotted line 200 is obtained by linearly interpolating the heartbeat interval after removing the outliers (spline interpolation or the like is also possible). , A method of resampling the interpolated value on the dotted line 200 at each cycle (for example, 0.5 seconds) of the reciprocal of the above resampling frequency, such as points 211, 212, ..., Point 223, can be mentioned. ..

Further, the acquisition unit 11 acquires the acceleration data string by sampling the detection value of the acceleration sensor at a certain sampling frequency (for example, 2 Hz). The sampling frequency and resampling frequency are not limited to the exemplified 250 Hz and 2 Hz. In particular, the sampling frequency of the pulse wave sensor, 250 Hz, is a value with a sufficient margin, and it is considered that there is no problem even if the sampling frequency is actually about 80 Hz to 100 Hz. Further, although the resampling frequency of the heartbeat interval and the sampling frequency of the acceleration illustrated above both match at 2 Hz, it is not necessary to match them.

Then, when estimating the sleep state at the reference time point t, the acquisition unit 11 uses the data string of the equidistant RRI and the data string of the acceleration as shown in FIG. Data of a certain number of samples (512, 256, 128, 64) corresponding to each time window is extracted from a plurality of long time windows (windows 1 to 19). In FIG. 5, the equidistant RRI data sequence is indicated by the dotted line 231 and the acceleration data sequence is indicated by the alternate long and short dash line 232. Hereinafter, the data string extracted in each time window from the data string of the equidistant RRI is represented by rdata [p] [i], and the data string extracted in each time window from the acceleration data string is mdata [p] [i]. ]. Note that p is the time window number and i is the sample number. That is, p is a value from 1 to 19, and if the number of samples in the time window used for extracting the data sequence rdata [p] [i] and the data sequence mdata [p] [i] is n, i is 1 to n. It becomes the value up to.

As shown in FIG. 5, the window 1 is, for example, a time window of 512 samples centered on the reference time point t, and as described above, the resampling frequency of the data string of the equidistant RRI and the accelerometer. The sampling frequency of each of the detected values of is 2 Hz, and the sampling period of these is 0.5 second. From the above, when the data strings are extracted from the window 1, 512 samples are extracted from each data string during a period of 0.5 seconds × 512 = 256 seconds = 4 minutes 16 seconds centered on the reference time point t. Data can be obtained. Further, the window 3 is a time window centered on the reference time t, for example, 256 samples, and when a data string is extracted through the window 3, 0.5, which is centered on the reference time t, is extracted from each data string. Data for 256 samples is obtained for a period of seconds x 256 = 128 seconds = 2 minutes and 8 seconds. For windows 2 and 4, for example, the number of samples is 256, which is the same as that of window 3, but the center timing is shifted back and forth by, for example, 128 samples (= 64 seconds). , Data for 256 samples can be obtained with the center timing shifted by 128 samples back and forth.

The windows 5 to 11 are, for example, 128 samples, and as shown in FIG. 5, are time windows arranged in this order from the front on the time axis. The window 5 at the front of these is a time window whose center timing is, for example, 128 + 64 samples (= 96 seconds) before the reference time t. Therefore, when the data string is extracted by the window 5, the reference time is extracted from each data string. Data for 128 samples can be obtained during a period of 0.5 seconds x 128 = 64 seconds = 1 minute and 4 seconds centered on t-96 seconds. The number of samples of windows 6 to 11 is 128, which is the same as that of window 5, but the center timing is shifted by, for example, 64 samples (= 32 seconds). Therefore, when the data string is extracted in these time windows, the center is obtained. Data for 128 samples can be obtained with the timing of the above shifted by 64 samples.

The windows 12 to 19 are, for example, 64 samples, and as shown in FIG. 5, they are time windows arranged in this order from the front on the time axis. The window 12 at the front of these is a time window whose center timing is, for example, 128 + 64 + 32 samples (= 112 seconds) before the reference time t. Therefore, when the data string is extracted from the window 12, the reference time is extracted from each data string. Data for 64 samples can be obtained for a period of 0.5 seconds x 64 = 32 seconds centered on t-112 seconds. The number of samples of windows 13 to 19 is 64, which is the same as that of window 12, but the center timing is shifted by, for example, 64 samples (= 32 seconds). Therefore, when the data string is extracted in these time windows, the center is obtained. Data for 64 samples can be obtained with the timing of the above shifted by 64 samples.

In the above description, each time window is a window having different time lengths based on the reference time point t, but it is not always necessary to use the reference time point t as a reference. That is, the time window may be a set of a plurality of time windows set in different periods within a certain period even if the reference time point t does not exist.

Next, the acquisition unit 11 calculates the feature amount (RRI system feature amount, body movement system feature amount, respiratory system feature amount) for each time window from the data sequence extracted by each time window in this way. .. As described above, assuming that the time window numbers (1 to 19) are represented by p and the sample numbers in the data strings extracted in each time window are represented by i, the data string of the equidistant RRI is rdata [. The data strings of p] [i] and acceleration can be expressed as mdata [p] [i].

The RRI feature amount is a feature amount obtained from time-series data of evenly spaced RRI, and more specifically, it is divided into a frequency system feature amount and a time system feature amount. The frequency system feature quantity is a feature quantity obtained from the frequency components of the data strings rdata [p] [i] of the equidistant RRI extracted in the above-mentioned time window. Further, the time-based feature amount is a feature amount obtained from the data sequence rdata [p] [i] (time-series data) itself of the equidistant RRI. The body movement feature amount described later is also a time system feature amount because it is a feature amount obtained from the acceleration data sequence mdata [p] [i] (time series data) itself extracted in the time window described above.

In the present embodiment, the following vlf, lf, hf, tf, hf_lfhf, lf_hf, and llf / tf are performed by performing FFT (Fast Fourier Transform) on rdata [p] [i] as frequency system features. , Lf / tf, hf / tf, 9 values are derived for each time window. However, each value is as follows.
vlf: Power spectrum below vlf_max (eg 0.01 Hz).
lf: A power spectrum greater than vlf_max and less than or equal to lf_max (eg 0.15 Hz).
hf: Power spectrum greater than lf_max and less than or equal to hf_max (eg 0.5 Hz).
tf = vlf + lf + hf
hf_lfhf = hf / (hf + lf)
lf_hf = lf / hf
Since the above nine values are obtained for each time window and there are 19 time windows, a total of 9 × 19 = 171 feature quantities are derived.

Further, in the present embodiment, as the time-based features of the RRI-based features, the following six mRRI, mHR, sdRRI, cvRRI, RMSSD, and pNN50 are used from the time-series data rdata [p] [i] of the equidistant RRI. The value of is derived for each time window. However, the unit of the value of rdata [p] [i] is milliseconds, and each value is as follows.
average value of mRRI = rdata [p] [i] mHR = 60000 / mRRI
(Note: In order to obtain the number of beats per minute (mHR) from mRRI in milliseconds, 60 seconds x 1000 = 60,000 is divided by mRRI.)
Standard deviation of sdRRI = rdata [p] [i] cvRRI = sdRRI / mRRI
RMSD = square root mean square of the difference between adjacent rdata [p] [i] pNN50 = ratio of the difference between adjacent rdata [p] [i] exceeding a predetermined time (for example, 50 milliseconds) (Note) : Since the rate at which the difference between consecutive adjacent heartbeat intervals exceeds 50 milliseconds is an index of vagal nerve tone intensity, 50 milliseconds is adopted as the predetermined time in this embodiment.)
Since the above 6 values are obtained for each time window and there are 19 time windows, a total of 6 × 19 = 114 feature quantities are derived.

The body movement system feature amount is a feature amount obtained from time-series data of detected values (acceleration values of each of the three axes) obtained by the acceleration sensor. In the present embodiment, the following five values of mACT, sdACT, minACT, maxACT, and cvACT are used as the body movement system features from the acceleration data sequence mdata [p] [i] extracted in the time window described above. , Derived for each time window. However, each value is as follows.
mACT = mean value of mdata [p] [i] sdACT = standard deviation of mdata [p] [i] minACT = minimum value of mdata [p] [i] maxACT = maximum value of mdata [p] [i] cvACT = sdACT / mACT
Since the above five values are obtained for each time window and there are 19 time windows, a total of 5 × 19 = 95 feature quantities are derived.

The respiratory feature amount is a feature amount related to respiration. A dedicated sensor for detecting respiration may be provided in the sensor unit 30 in order to obtain this feature amount, but the time-series data of the equidistant RRI includes pulse wave respiratory sinus arrhythmia, which is a variable rhythm depending on respiration. (Respiratory Sinus Arrhythmia: RSA) is included, and this is used in this embodiment. Specifically, the time-series data of the equidistant RRI is extracted in the time window (window 2, window 3 and window 4) having the above-mentioned sample number of 256, and extracted in each time window (window 2, window 3 and window 4). FFT is performed on the resulting data strings (rdata [2] [i], rdata [3] [i] and rdata [4] [i]), and for example, 0.1 Hz to 0.4 Hz (LF (Low Frequency) band). ) Is found, and the frequency at which the energy is maximized is determined as RSA [j]. However, j corresponds to the time window used for extraction (window 2: j = 0, window 3: j = 1, window 4: j = 2).

Then, the following six values of mRSA, sdRSA, minRSA, maxRSA, cvRSA, and RSA [1] obtained from the RSA [j] thus obtained are used. However, each value is as follows.
mRSA = average value of RSA [j] = (RSA [0] + RSA [1] + RSA [2]) / 3
Standard deviation of sdRSA = RSA [j] = Minimum value of standard deviation minRSA = RSA [j] obtained from RSA [0], RSA [1] and RSA [2] = min (RSA [0], RSA [1]] , RSA [2])
max RSA = RSA [j] maximum value = max (RSA [0], RSA [1], RSA [2])
cvRSA = sdRSA / mRSA (However, if mRSA≈0, cvRSA = 0)
RSA [1] = RSA obtained from the data string extracted in window 3, which is the central time window.
As described above, since the values of the respiratory system features cannot be obtained for each time window, the above six features are derived. As for the respiratory feature amount, a large number of feature amounts may be derived by calculating the value for each time window.

In this way, the acquisition unit 11 extracts the time-series data of biological signals (equally spaced RRI and acceleration) in a plurality of time windows having different periods from each other, and uses the extracted time-series data to obtain RRI-based features. Acquire features such as body movement features and respiratory features. In the present embodiment, since the above-mentioned reference time point t is advanced by a certain time (for example, 30 seconds), the acquisition unit 11 acquires these feature quantities at each certain time. The above-mentioned feature amount is an example, and the acquisition unit 11 may acquire only a part of the above-mentioned feature amount, or may acquire a feature amount other than the above-mentioned feature amount. Good.

The determination unit 12 determines whether or not the first REM sleep state has appeared since the subject entered the bed as the subject's sleep state. However, normally, during sleep, all sleep states including the REM sleep state appear in one cycle (about 90 to 120 minutes). Therefore, in the present embodiment, the determination unit 12 determines that the first REM sleep state has appeared because a predetermined time (for example, 120 minutes) has elapsed since the subject entered the bed. How to determine the subject's entry into the floor is arbitrary, but in the present embodiment, when the acceleration sensor detects that the subject has been lying down continuously for a certain period of time or longer, the subject has entered the bed. Deemed, the time when the subject lay down is determined to be the time of bedtime (the time when the subject entered the bed). The determination unit 12 functions as a determination means. Here, entering the bed does not necessarily mean that the subject lies on a bed or the like. This includes taking a posture for the subject to sleep, such as sitting on a chair for the subject to sleep, and performing an action for sleeping.

The determination unit 12 may determine whether or not a certain sleep state other than the REM sleep state (for example, any of the above-mentioned walk, N1, N2, and N3) has appeared. As described above, since all sleep states usually appear in one cycle (about 90 to 120 minutes) during sleep, the determination unit 12 determines a predetermined time (for example, 120 minutes) after the subject enters the bed. ) Has passed, and it may be determined that the certain sleep state has appeared.

The generation unit 13 is a biological signal from the time the subject enters the bed to the time when the determination unit 12 determines that a REM sleep state (or a certain sleep state) has appeared (when two hours have passed since the subject entered the bed). A standardization parameter is generated based on the feature amount (RRI system feature amount, body movement system feature amount, respiratory system feature amount, etc.) acquired by the acquisition unit 11 using the above. In the present embodiment, standardization parameters are generated for each of the RRI system feature amount, the body movement system feature amount, and the respiratory system feature amount. Further, in the present embodiment, the standardization parameter is defined as the time when the subject enters the bed is t = t [0], and the time when the determination unit 12 determines that the REM sleep state appears is t = t [z]. , The mean value and standard deviation of the feature amount from t = t [0] to t = t [z].

Specifically, the data of a certain feature amount is x [i] (where i is an integer from 1 to n. The feature amount at the timing t = t [0] is x [1], and t = t [z]. The feature amount in (1) and (2) is expressed by (1) and (2) below, and the generation unit 13 calculates the mean value a and the standard deviation s of the feature amount x [i] by the following equations (1) and (2). Calculated as a standardization parameter.

The generation unit 13 functions as a generation means.

Based on the standardization parameters generated (calculated) by the generation unit 13, the standardization unit 14 obtains the feature amounts acquired by the acquisition unit 11 (for each of the RRI system feature amount, the body movement system feature amount, and the respiratory system feature amount). )Standardize. Specifically, assuming that the data of a certain feature amount is represented by x [i] as described above, z [i] is calculated as a standardized feature amount by the following equation (3).
z [i] = (x [i] -a) / s ... (3)
By doing so, the standardized features have an average value of 0 and a variance of 1. The standardization unit 14 functions as a standardization means.

The estimation unit 15 estimates the sleep state of the subject based on the features standardized by the standardization unit 14. The estimation unit 15 functions as an estimation means. For this estimation, a pre-learned "multi-class-" is used to output a parameter representing the sleep state (one of the five states of walk, REM, N1, N2, and N3) when a standardized feature is input. Use "Logistic Regression". This is a two-class classifier that identifies whether the subject's sleep state is work or another sleep state, a two-class classifier that discriminates between REM and other sleep states, A two-class classifier that identifies whether N1 or other sleep states, a two-class classifier that identifies N2 or other sleep states, N3 and other sleep states Five two-class classifiers, which are two-class classifiers that identify which of the two, are prepared, and the two classes that maximize the output value when standardized feature quantities are input to each of these five classifiers. The sleep state corresponding to the discriminator is estimated to be the sleep state of the subject. If a plurality of two-class classifiers having the maximum output value are generated, the sleep state most recently estimated to be the sleep state is used for estimation.

The data of the subject to which the correct answer data of the sleeping state is labeled is prepared as the learning data at a certain time (30 seconds) for advancing the reference time point t based on the brain wave or the like by an expert. The classifier is pre-learned using this learning data. The learning of these classifiers is repeated using a number of subjects, and for each subject, the feature amount is acquired from the biological signal of the subject overnight, and the standardization parameter (mean of the feature amount) is obtained from the acquired feature amount. (Value and standard deviation) are obtained, the features are standardized using the obtained standardization parameters, and the classifier is trained using the standardized features.

The functional configuration of the sleep state estimation device 100 has been described above. Next, the sleep state estimation process of the sleep state estimation device 100 will be described with reference to FIG. When the sleep state estimation device 100 receives an instruction from the user to start the sleep state estimation process via the input unit 41, the sleep state estimation device 100 starts the sleep state estimation process. The user of the sleep state estimation device 100 may or may not be a subject of the sleep state estimation process.

First, the acquisition unit 11 of the sleep state estimation device 100 acquires biological signals (pulse wave, acceleration, etc.) detected by the sensor included in the sensor unit 30 (step S101). The first execution of step S101 is performed for a time equal to or longer than the time of the above-mentioned window 1 (for example, 5 minutes), and the second and subsequent steps S101 set the reference time point t on the time axis by a predetermined time (for example, 30 seconds). It is done while shifting. That is, at the first execution of step S101, a biological signal sufficient for data string extraction in windows 1 to 19 is acquired, and from the second time onward, biological signals for the next predetermined time (for example, 30 seconds) are acquired. .. When the predetermined time is 30 seconds, the data sampled at 100 Hz such as pulse wave is 30,000 data strings, and the data sampled at 2 Hz such as acceleration is 60 data in this 30 seconds. Each is acquired as a data string and stored in the storage unit 20.

Next, the acquisition unit 11 calculates a beat interval data string along the time series from the pulse wave data string among the biometric signals stored in the storage unit 20, and recycles the beat interval data string at, for example, 2 Hz. By sampling, a data string (rdata [p] [i]) of evenly spaced RRIs every 0.5 seconds is calculated (step S102).

Next, the acquisition unit 11 generates a data string of body motion data (mdata [p] [i]) from the acceleration data string among the biological signals stored in the storage unit 20 (step S103). Each data included in the acceleration data string is a triplet of data (AX (acceleration in the X-axis direction), AY (acceleration in the Y-axis direction), AZ (acceleration in the Z-axis direction)). , Body movement data is the norm of acceleration obtained from this triplet of data (AX, AY, AZ). Therefore, mdata [p] [i] is expressed by the following equation (4).

Next, the acquisition unit 11 uses the feature amount (RRI system feature amount, body movement) from the equidistant RRI data sequence (rdata [p] [i]) and the body movement data data sequence (mdata [p] [i]). Each of the system feature amount and the respiratory system feature amount) is calculated and stored in the storage unit 20 (step S104). The calculation method of each feature amount is as described above. Step S104 is also called an acquisition step.

Next, the determination unit 12 determines whether or not two hours, which is a predetermined time, has elapsed since the subject entered the bed (step S105). Step S105 is also called a determination step. This is because it is usually considered that all sleep states appear two hours after entering the bed. If 2 hours have not passed since the subject entered the bed (step S105; No), the process returns to step S101. If two hours have passed since the subject entered the bed (step S105; Yes), the generation unit 13 generates standardization parameters as described above based on the feature quantities stored in the storage unit 20 up to that point. (Calculate) (step S106). Step S106 is also called a generation step.

Then, the standardization unit 14 uses the standardization parameters to standardize the feature amount stored in the storage unit 20 as described above (step S107). Step S107 is also called a standardization step. Then, the estimation unit 15 estimates the sleep state from entering the bed using the standardized features (step S108). The sleep state estimation method is as described above. Step S108 is also called an estimation step.

Next, the acquisition unit 11 acquires a biological signal (pulse wave, acceleration, etc.) detected by the sensor included in the sensor unit 30 (step S109). Step S109 is also performed while shifting the reference time point t by a predetermined time (for example, 30 seconds) in the same manner as in the second and subsequent steps S101 described above.

The processing of the following steps S110, step S111, step S112, and step S113 is the same as the processing of step S102, step S103, step S104, and step S107, respectively. Then, the estimation unit 15 estimates the sleep state as described above using the standardized features (step S114).

Then, the control unit 10 determines whether or not the end condition is satisfied (step S115). Any condition can be set as the end condition. For example, it may end when the time reaches 7:00 in the morning, it may end when the user inputs an end instruction from the input unit 41, or the estimation result by the estimation unit 15 is in a predetermined state (for example, work). You may end it when it becomes. Further, a combination of these conditions (for example, a combination of a time condition and an estimation result condition) may be used as the end condition. For example, if the sleep state (for example, work or REM) in which the estimation result by the estimation unit 15 is available within one hour before and after the time set by the subject as the alarm clock, the operation may be terminated.

If the end condition is not satisfied (step S115; No), the process returns to step S109. If the end condition is satisfied (step S115; Yes), the sleep state estimation process is terminated. When the sleep state estimation device 100 is used as an alarm clock, when the end condition is satisfied, the control unit 10 makes a sound from the output unit 42 or makes a vibrator before ending the sleep state estimation process. It may be vibrated.

By the above sleep state estimation process, after 2 hours have passed since the subject entered the bed, the sleep state estimation unit 15 can estimate the sleep state using the feature amount standardized by the standardization unit 14, so that the sleep state estimation device 100 Can estimate the sleep state in real time even when the subject is sleeping. Then, since it can be assumed that all sleep states (wake, REM, N1, N2, N3) have appeared between the time when the subject enters the bed and the time when two hours have passed, the generation unit 13 has all the sleep. It is possible to generate a standardization parameter after the state appears, and it is possible to improve the estimation accuracy of the sleep state by the estimation unit 15.

(Modification example 1)
In the sleep state estimation method (FIG. 6) according to the first embodiment described above, the standardization parameter generated in step S106 will continue to be used forever, but as the acquired features increase, the standardization parameter will be used. It is also possible to update. Such a modification 1 will be described.

The generation unit 13 according to the first modification is when the determination unit 12 determines that a REM sleep state (or a certain sleep state) has appeared after the subject has entered the bed (when two hours have passed since the subject entered the bed). ), The standardization parameters are generated in the same manner as in the first embodiment. However, after that, the generation unit 13 according to the modification 1 recalculates the mean value and the standard deviation of the feature amount by using all the feature amounts acquired so far as the feature amount is acquired. The recalculated mean value and standard deviation are used as standardization parameters. That is, the generation unit 13 according to the modification 1 updates the standardization parameters as the acquisition unit 11 acquires the feature amount.

The sleep state estimation process of the sleep state estimation device 100 according to the first modification will be described with reference to FIG. 7. Since this process is the same process as the sleep state estimation process (FIG. 6) according to the first embodiment except for a part, the differences will be mainly described.

Since the processes from step S101 to step S112 are the same as the sleep state estimation process (FIG. 6) according to the first embodiment, the description thereof will be omitted. After the acquisition unit 11 calculates the feature amount in step S112, the generation unit 13 updates the standardization parameter using the feature amount obtained so far (step S121). Then, since the processing after step S113 after step S121 is the same processing as the sleep state estimation processing (FIG. 6) according to the first embodiment, the description thereof will be omitted.

By the above sleep state estimation process, as in the case of the sleep state estimation process according to the first embodiment, the sleep state can be estimated in real time even when the subject is sleeping, and the sleep in the estimation unit 15 can be estimated. The accuracy of state estimation can be improved. Further, since the sleep state estimation device 100 according to the first modification continues to update the standardization parameters even after 2 hours have passed since the subject entered the bed, the accuracy of estimating the sleep state by the estimation unit 15 is further improved. be able to.

(Modification 2)
In the sleep state estimation method (FIG. 6) according to the first embodiment and the sleep state estimation method (FIG. 7) according to the first modification, in step S105, whether or not two hours have passed since the subject entered the bed. Was judged. This is because it is usually considered that all sleep states appear 2 hours after entering the bed. However, as an exception, all sleep states may appear before the lapse of 2 hours, or some sleep states may not appear even after 2 hours have passed. However, it is usually believed that all sleep states appear between the time of bed entry and the appearance of the first REM sleep state. Therefore, instead of determining whether or not two hours have passed since entering the bed, it is conceivable to determine whether or not the REM sleep state appears for the first time after entering the bed. Such a modification 2 will be described.

The determination unit 12 according to the second modification uses a two-class classifier learned in advance so as to output whether the sleep state is the REM sleep state or the other sleep state when the feature amount is input. Therefore, it is determined whether or not the REM sleep state appears without using the passage of time (120 minutes). This two-class classifier discriminates using only the features that can be identified with a certain degree of accuracy without standardizing the features by the standardization unit 14. The "features that can be identified with a certain degree of accuracy without standardizing the features" are the features that use the power ratio of each frequency band among the frequency features and the statistics of the time features. It is a feature quantity using quantities (coefficient of variation and standard deviation).

Specifically, the determination unit 12 of the modified example 2 uses the above-mentioned vlf / tf, lf / tf, and hf / tf as feature amounts using the power ratio of each frequency band among the frequency system feature amounts. Further, the determination unit 12 of the modification 2 uses the above-mentioned cvACT and sdACT as statistics (coefficient of variation and standard deviation) among the time-based features. Similar to the case of the first embodiment, when learning the two-class classifier provided in the determination unit 12 of the modified example 2, the data of the subject to which the correct answer data of the sleeping state is labeled by the expert based on the brain waves and the like. Is used as learning data. Then, a number of subjects are used to repeatedly learn. However, unlike the case of the first embodiment, the feature amount acquired from the data of this subject is used as it is (without standardization) for learning.

The sleep state estimation process of the sleep state estimation device 100 according to the second modification will be described with reference to FIG. Since this process is the same as the sleep state estimation process (FIG. 6) and the sleep state estimation process (FIG. 7) according to the first embodiment except for a part, the differences are mainly focused on. Explain to.

Since the processes from step S101 to step S104 are the same as the sleep state estimation process (FIG. 6) and the sleep state estimation process (FIG. 7) according to the first embodiment, the description thereof will be omitted. To do. After the acquisition unit 11 calculates the feature amount in step S104, the determination unit 12 determines whether or not the REM sleep state has appeared (step S131). Step S131 is also called a determination step. If the REM sleep state has not yet appeared (step S131; No), the process returns to step S101. When the REM sleep state appears (step S131; Yes), the process proceeds to step S106, and the generation unit 13 generates a standardization parameter based on the feature amount stored in the storage unit 20 so far. Then, since the processing after step S106 is the same processing as the sleep state estimation processing (FIG. 7) according to the first modification, the description thereof will be omitted.

In FIG. 8, the generation unit 13 performs the process of updating the standardization parameter in step S121 as in the modification 1 (FIG. 7), but this process is omitted and the first embodiment (FIG. 6). ), The standardization parameter generated in step S106 may continue to be used.

After estimating the first REM sleep state after the subject enters the bed by the above sleep state estimation process, the estimation unit 15 can estimate the sleep state using the feature amount standardized by the standardization unit 14. The sleep state estimation device 100 according to the second modification can estimate the sleep state in real time even when the subject is sleeping. Then, since it can be assumed that all sleep states (wake, REM, N1, N2, N3) have appeared between the time when the subject enters the bed and the time when the REM sleep state appears, the generation unit 13 has all the sleep states. It is possible to generate a standardization parameter after the sleep state appears, and it is possible to improve the estimation accuracy of the sleep state by the estimation unit 15. Further, the sleep state estimation device 100 according to the modification 2 continues to update the standardization parameters even after the subject enters the bed and the REM sleep state appears, as in the case of the modification 1, so that the estimation unit 15 It is possible to further improve the estimation accuracy of the sleep state in.

The determination unit 12 according to the second modification used a two-class classifier learned in advance so as to output a REM sleep state or a sleep state other than the REM sleep state when the feature amount is input. However, the sleep state identified by the two-class classifier does not have to be the REM sleep state. If it is a sleep state that can be identified using only the feature amount that can be identified with a certain degree of accuracy without standardizing the feature amount by the standardization unit 14, any sleep state is identified by this two-class classifier. be able to. In this case, the determination unit 12 can be used as a determination means for determining whether or not a certain sleep state has appeared.

(Modification example 3)
Exceptionally, it is conceivable that all sleep states do not appear even after 2 hours, which is a predetermined time, from entering the bed, or that all sleep states do not appear until the REM sleep state appears. If the determination unit 12 determines that "two hours have passed since entering the bed" and "a REM sleep state appears at least once", such an exceptional case can be dealt with. Such a modification 3 will be described.

The determination unit 12 according to the modified example 3 determines that two hours have passed since entering the bed, and similarly to the determination unit 12 according to the modified example 2, when the feature amount is input, the sleep state becomes REM sleep. By using a pre-learned two-class classifier to output whether it is a state or another sleep state, it is determined whether or not a REM sleep state has appeared. Then, when two hours have passed since entering the bed and the two-class classifier outputs the REM sleep state at least once, it is determined that the REM sleep state has appeared as the subject's sleep state.

The sleep state estimation process of the sleep state estimation device 100 according to the third modification is the same as the sleep state estimation process (FIG. 8) of the sleep state estimation device 100 according to the second modification, except for step S131. In the sleep state estimation process according to the third modification, in step S131, the determination unit 12 "two hours have passed since entering the bed, and the two-class classifier used by the determination unit 12 by then determines the REM sleep state. It is determined whether or not the output has been performed at least once. If two hours have not passed since entering the bed and if the REM sleep state is not output from the two-class classifier (step S131; No), the process returns to step S101. If two hours have passed since entering the bed and the REM sleep state appears one or more times (step S131; Yes), the process proceeds to step S106.

Since the processes other than step S131 are the same as the sleep state estimation process (FIG. 8) according to the second modification, the description thereof will be omitted.

After 2 hours have passed since the subject entered the bed by the above sleep state estimation process and the REM sleep state was estimated at least once, the estimation was performed using the feature amount standardized by the standardization unit 14. Since the unit 15 can estimate the sleep state, the sleep state estimation device 100 according to the modified example 3 can estimate the sleep state in real time even when the subject is sleeping. And even in exceptional cases, all sleep states (wake) until 2 hours have passed since the subject entered the bed and the REM sleep state is output from the 2-class classifier. , REM, N1, N2, N3) can be assumed to have appeared, so that the generation unit 13 can generate the standardization parameters after all the sleep states have appeared, and the estimation unit 15 estimates the sleep state. The accuracy can be improved.

(Other variants)
In the above-described embodiment and modification, the pulse wave sensor for detecting the pulse wave and the acceleration sensor for detecting the movement of the body are provided, but the type of the sensor included in the sensor unit 30 is not limited to these. The sensor unit 30 can be provided with any sensor as long as it is a sensor that detects (acquires) a biological signal for estimating a sleep state. Further, when the biological signal or the feature amount for estimating the sleep state can be received from an external device or the like via, for example, the communication unit 43, the sleep state estimation device 100 does not need to include the sensor unit 30.

In the above-described embodiment and modified example, the sleep state estimation device 100 uses the pulse wave by the pulse wave sensor attached to the ear and the acceleration by the acceleration sensor attached to the ear as the biological signal. , Biological signals are not limited to these. The biological signals that can be used by the sleep state estimation device 100 include body movement (detected by acceleration sensors installed on the head, arms, chest, legs, torso, etc.) and myoelectricity (head (around the temples and nose)). , Detected by myoelectric sensors installed on the arms, chest, legs, torso, etc.), Sweat (detected by skin electrometer and humidity sensor), Heart rate (Electrocardiograph, pressure sensor installed under the bed (cardiac motion diagram) Waveform detection), detection by pulse wave sensors installed on the head, arms, chest, legs, torso, etc.), etc. are included.

Further, in the above-described embodiment and modified example, the sleep state estimation device 100 estimates the sleep state of a human subject. However, the target for estimating the sleep state is not limited to humans, and it is possible to target animals in general such as dogs, cats, horses, cows, pigs, and chickens. This is because it is possible to attach a pulse wave sensor or an acceleration sensor to these animals to acquire the features required for estimating the sleep state.

Further, in the above-described embodiment and modification, the estimation unit 15 of the sleep state estimation device 100 uses multi-class-logistic regression, but the discriminator used by the estimation unit 15 is not limited to multi-class-logistic regression. For example, the estimation unit 15 may use other classifiers such as SVC (Support Vector Machine Classification) and SVR (Support Vector Regression).

Similarly, a two-class classifier used by the determination unit 12 according to the second modification (a two-class classifier that outputs whether the sleep state is a REM sleep state or another sleep state when a feature amount is input) is also available. , 2 Class-Logistic Regression, SVC, SVR, etc., any classifier can be used.

Further, in the above-described embodiment and modified example, the sleep state estimation device 100 uses the RRI system feature amount, the body movement system feature amount, and the respiratory system feature amount as the feature amount for performing the sleep estimation. The amount is not limited to the above-mentioned feature amount, and any feature amount can be used as long as it is a feature amount for estimating sleep. For example, when myoelectricity, sweat, etc. are also used as biological signals, the detected values for these are sampled at, for example, 2 Hz in the same manner as other feature quantities, and data are obtained in a plurality of time windows with reference to the reference time point t. May be extracted to calculate the feature amount.

Further, in the above-described embodiment and modification, the estimation unit 15 estimates the sleep state using all the feature amounts acquired by the acquisition unit 11, but it is effective to use, for example, RFE (Recursive Feature Elimination) or the like. It is also possible to extract only the feature amount and estimate the sleep state using only the extracted effective feature amount.

Further, in the above-described embodiment and modification, the generation unit 13 uses the average value and standard deviation of the feature amount data string as the standardization parameters. However, the parameters for standardization are not limited to these. Any standardization parameter can be used as long as it is a parameter for standardizing the feature quantity. For example, the intermediate value or the mode value of the data string may be used. Also, the "standardization" itself is not limited to general standardization (the mean value of data is set to 0 and the variance is set to 1). For example, processing that performs some operation in a unified manner on the feature amount (biological signal parameter), such as reducing the feature amount of frequently occurring data and increasing the feature amount of data that appears rarely, is included in standardization.

Further, in the above-described embodiment and modification, the acquisition unit 11 acquires the feature amount calculated from the biological signal, the generation unit 13 generates the standardization parameter based on the feature amount, and the standardization unit 14 generates the standardization parameter. The features were standardized based on. However, the object of standardization is not limited to the feature amount, and may be the biological signal itself. In this case, the acquisition unit 11 acquires the biological signal, the generation unit 13 generates the standardization parameter based on the biological signal, and the standardization unit 14 standardizes the biological signal based on the standardization parameter. Then, the estimation unit 15 calculates a feature amount from the standardized biological signal, and estimates the sleep state based on the calculated feature amount.

Furthermore, the biological signal itself may be standardized for a part of the biological signals, and the feature amount may be calculated and then standardized for the other parts. In this case, the estimation unit 15 estimates the sleep state based on the standardized feature amount and the feature amount calculated from the standardized biological signal. As described above, the object of standardization may be the biological signal itself or the feature amount calculated from the biological signal. Therefore, those to be standardized are collectively referred to as biological signal parameters. The biological signal parameters include both the biological signal itself and the feature amount calculated from the biological signal.

In the above-described embodiment, the sleep state estimation device 100 includes an input unit 41, an output unit 42, and a communication unit 43, but these are not essential components, and the sleep state estimation device 100 includes these. It does not have to be.

Further, each function of the sleep state estimation device 100 can also be performed by a computer such as a normal PC (Personal Computer). Specifically, in the above embodiment, the program such as the sleep state estimation process performed by the sleep state estimation device 100 has been described as being stored in the ROM of the storage unit 20 in advance. However, the program can be a flexible disk, a CD-ROM (Compact Disk Read Only Memory), a DVD (Digital Versaille Disc), an MO (Magnet-Optical Disc), a memory card, a USB (Universal Serial Bus) memory, or the like. A computer capable of realizing each of the above-mentioned functions may be configured by storing and distributing the program in a recording medium, reading the program into a computer, and installing the program.

Although the preferred embodiment of the present invention has been described above, the present invention is not limited to the specific embodiment, and the present invention includes the invention described in the claims and the equivalent range thereof. Is done. The inventions described in the claims of the original application of the present application are described below.

(Appendix 1)
An acquisition means for acquiring biological signal parameters based on the target biological signal, and
A determination means for determining whether or not a certain sleep state has appeared as the target sleep state, and
A generation means for generating standardization parameters based on the biological signal parameters up to the time when it is determined by the determination means that a certain sleep state has appeared as the target sleep state.
A standardization means for standardizing the biological signal parameter based on the generated standardization parameter,
An estimation means for estimating the sleep state of the subject based on the standardized biological signal parameters, and
A sleep state estimator equipped with.

(Appendix 2)
The biological signal parameter is a feature amount calculated from the biological signal.
The sleep state estimation device according to Appendix 1.

(Appendix 3)
The acquisition means
Time-series data of the biological signal is extracted in a plurality of time windows having different time lengths.
The feature amount is calculated using the extracted time series data.
The sleep state estimation device according to Appendix 2.

(Appendix 4)
The determination means determines that a certain sleep state has appeared as the sleep state of the target when a predetermined time has elapsed since the subject entered the bed.
The sleep state estimation device according to any one of Supplementary note 1 to 3.

(Appendix 5)
The determination means determines whether or not a certain sleep state appears as the sleep state of the target based on the acquired biological signal parameters.
The sleep state estimation device according to any one of Supplementary note 1 to 4.

(Appendix 6)
The certain sleep state is REM sleep state,
The sleep state estimation device according to any one of Supplementary note 1 to 5.

(Appendix 7)
Even after it is determined by the determination means that a certain sleep state has appeared as the target sleep state, the generation means updates the standardization parameter based on the biological signal parameters acquired so far.
The standardization means standardizes the biological signal parameters based on the updated standardization parameters.
The sleep state estimation device according to any one of Supplementary note 1 to 6.

(Appendix 8)
The subject is a human
The biological signals of the subject are pulse waves and body movements.
The sleep state estimation device according to any one of Supplementary note 1 to 7.

(Appendix 9)
The generation means generates the average value and standard deviation of the biological signal parameters as standardization parameters.
The standardization means standardizes the biological signal parameter by dividing the value obtained by subtracting the mean value from the biological signal parameter by the standard deviation.
The sleep state estimation device according to any one of Supplementary note 1 to 8.

(Appendix 10)
The acquisition step to acquire the biological signal parameters based on the target biological signal,
A determination step for determining whether or not a certain sleep state has appeared as the target sleep state, and
A generation step of generating standardization parameters based on the biological signal parameters up to the time when it is determined that a certain sleep state has appeared as the target sleep state by the determination step.
A standardization step for standardizing the biological signal parameters based on the generated standardization parameters, and
An estimation step of estimating the sleep state of the subject based on the standardized biological signal parameters, and
A sleep state estimation method that comprises.

(Appendix 11)
On the computer
Acquisition step to acquire biological signal parameters based on the biological signal of the target,
A determination step for determining whether or not a certain sleep state has appeared as the target sleep state,
A generation step of generating standardization parameters based on the biological signal parameters up to the time when it is determined that a certain sleep state has appeared as the target sleep state by the determination step.
A standardization step for standardizing the biological signal parameters based on the generated standardization parameters, and
An estimation step that estimates the sleep state of the subject based on the standardized biological signal parameters.
A program to execute.

10 ... control unit, 11 ... acquisition unit, 12 ... judgment unit, 13 ... generation unit, 14 ... standardization unit, 15 ... estimation unit, 20 ... storage unit, 30 ... sensor unit, 41 ... input unit, 42 ... output unit, 43 ... Communication unit, 100 ... Sleep state estimation device, 200, 231 ... Dotted line, 201 ... Pulse wave, 202i, 203i, 204i, 205i, 206i, 207i ... Heart rate interval, 201t, 202t, 203t, 204t, 205t, 206t, 207t ... Beat timing, 211,212,213,214,215,216,217,218,219.220,221,222,223 ... Point, 232 ... One-dot chain line

Claims (11)

An acquisition means for acquiring biological signal parameters based on the target biological signal, and
A determination means for determining whether or not a certain sleep state has appeared as the target sleep state, and
A generation means for generating standardization parameters based on the biological signal parameters up to the time when it is determined by the determination means that a certain sleep state has appeared as the target sleep state.
A standardization means for standardizing the biological signal parameter based on the generated standardization parameter,
An estimation means for estimating the sleep state of the subject based on the standardized biological signal parameters, and
A sleep state estimator equipped with. The biological signal parameter is a feature amount calculated from the biological signal.
The sleep state estimation device according to claim 1. The acquisition means
Time-series data of the biological signal is extracted in a plurality of time windows having different time lengths.
The feature amount is calculated using the extracted time series data.
The sleep state estimation device according to claim 2. The determination means determines that a certain sleep state has appeared as the sleep state of the target when a predetermined time has elapsed since the subject entered the bed.
The sleep state estimation device according to any one of claims 1 to 3. The determination means determines whether or not a certain sleep state appears as the sleep state of the target based on the acquired biological signal parameters.
The sleep state estimation device according to any one of claims 1 to 4. The certain sleep state is REM sleep state,
The sleep state estimation device according to any one of claims 1 to 5. Even after it is determined by the determination means that a certain sleep state has appeared as the target sleep state, the generation means updates the standardization parameter based on the biological signal parameters acquired so far.
The standardization means standardizes the biological signal parameters based on the updated standardization parameters.
The sleep state estimation device according to any one of claims 1 to 6. The subject is a human
The biological signals of the subject are pulse waves and body movements.
The sleep state estimation device according to any one of claims 1 to 7. The generation means generates the average value and standard deviation of the biological signal parameters as standardization parameters.
The standardization means standardizes the biological signal parameter by dividing the value obtained by subtracting the mean value from the biological signal parameter by the standard deviation.
The sleep state estimation device according to any one of claims 1 to 8. The acquisition step to acquire the biological signal parameters based on the target biological signal,
A determination step for determining whether or not a certain sleep state has appeared as the target sleep state, and
A generation step of generating standardization parameters based on the biological signal parameters up to the time when it is determined that a certain sleep state has appeared as the target sleep state by the determination step.
A standardization step for standardizing the biological signal parameters based on the generated standardization parameters, and
An estimation step of estimating the sleep state of the subject based on the standardized biological signal parameters, and
A sleep state estimation method that comprises. On the computer
Acquisition step to acquire biological signal parameters based on the biological signal of the target,
A determination step for determining whether or not a certain sleep state has appeared as the target sleep state,
A generation step of generating standardization parameters based on the biological signal parameters up to the time when it is determined that a certain sleep state has appeared as the target sleep state by the determination step.
A standardization step for standardizing the biological signal parameters based on the generated standardization parameters, and
An estimation step that estimates the sleep state of the subject based on the standardized biological signal parameters.
A program to execute.

Images