Learning device, learning method, and measuring device

Abstract

The present invention relates to a learning device, a learning method, and a measuring device. The subject of the present invention is to more effectively learn the relationship between feature points in time series data. A learning device is provided, the learning device having a learning unit, the learning unit using first sensor data as learning data and training data based on second sensor data to learn outputs related to object feature points, the first sensor data being obtained by a first method and having a time length corresponding to a repetitive interval that can be periodically observed along the progression of time, the second sensor data being obtained by a second method in which the influence of noise is smaller than that of the first method, at a time when a predetermined time has passed from the start time of the time length related to the first sensor data, the object feature point being an object to be observed in the repetitive interval, and the predetermined time being set based on the time length from the start time of the repetitive interval until the time when the object feature point is expected to appear.

Description

Technical Field

[0001] The present invention relates to a learning device, a learning method, and a measuring device. Background Art

[0002] In recent years, technologies that use machine learning for identification and estimation have been developed. For example, Patent Document 1 discloses a learning-related technology for predicting the future values of time series data. This technology makes it possible to predict the value that data acquired in a time series will represent at any point in the future.

[0003] Patent Document 1: Japanese Patent Application Laid-Open No. 2001-325582

[0004] However, the technology described in Patent Document 1 uses future values at a target prediction time as training data for learning. In this case, it is difficult to learn features such as the transition of time series data after the future values. Summary of the Invention

[0005] Therefore, the present invention has been made in view of the above-mentioned problems, and an object of the present invention is to provide a technology capable of more efficiently learning the relationship between feature points in time series data.

[0006] In order to solve the above-mentioned problems, according to one aspect of the present invention, a learning device is provided, wherein the learning device includes a learning unit, wherein the learning unit uses first sensor data as learning data and training data based on second sensor data to learn outputs related to object feature points, wherein the first sensor data is obtained by a first method and has a time length corresponding to a repetitive interval that can be periodically observed as time progresses, and the second sensor data is data obtained by a second method in which the influence of noise is smaller than that of the first method, at a moment when a specified time has passed from the start moment of the above-mentioned time length related to the above-mentioned first sensor data, the above-mentioned object feature point is an object to be observed in the above-mentioned repetitive interval, and the above-mentioned specified time is set based on the time length from the start moment of the above-mentioned repetitive interval to the moment when the above-mentioned object feature point is expected to appear.

[0007] In addition, in order to solve the above-mentioned problem, according to another aspect of the present invention, a learning method is provided, which includes: using first sensor data as learning data, the first sensor data being obtained by a first method and having a time length corresponding to a repetitive interval that can be periodically observed as time progresses, using training data based on second sensor data, the second sensor data being obtained by a second method in which the influence of noise is smaller than that of the first method, and learning the output related to the object feature point that is the object of observation in the repetitive interval as data obtained at a moment after a specified time has passed from the start moment of the above-mentioned time length involved in the above-mentioned first sensor data, wherein the above-mentioned specified time is set based on the length of time from the start moment of the above-mentioned repetitive interval to the moment when the above-mentioned object feature point is expected to appear.

[0008] In addition, in order to solve the above-mentioned problem, according to another aspect of the present invention, a measuring device is provided, wherein the measuring device includes a measuring unit, which takes first sensor data obtained by a first method as input and performs measurements related to object feature points that are an object to be observed in the first sensor data. The measuring unit uses a learned model to perform measurements related to the object feature points. The learned model uses the first sensor data having a time length corresponding to a repetitive interval that can be periodically observed along the progression of time as learning data, and learns the output related to the object feature points in the repetitive interval using training data based on second sensor data. The second sensor data is data obtained at a time when a specified time has passed from the start time of the time length related to the first sensor data using a second method in which the influence of noise is smaller than that of the first method, and the specified time is set based on the time length from the start time of the repetitive interval to the time when the object feature point is expected to appear.

[0009] As described above, according to the present invention, a technique capable of more efficiently learning the relationship between feature points in time-series data is provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a diagram showing a functional configuration example of a learning device 10 according to one embodiment of the present invention.

[0011] Figure 2 It is a diagram showing an example of the functional configuration of the measurement device 20 according to this embodiment.

[0012] Figure 3 This is a diagram showing an example of a general electrocardiographic waveform in one cycle.

[0013] Figure 4This is a diagram showing an example of correspondence between learning data and training data according to one embodiment of the present invention.

[0014] Figure 5 1 and 2 are diagrams showing an image of object feature points measured by the measurement unit 220 according to this embodiment.

[0015] Figure 6 1 and 2 are diagrams showing an image of object feature points measured by the measurement unit 220 according to this embodiment.

[0016] Figure 7 It is a graph showing the R-wave detection accuracy using the learned model according to this embodiment.

[0017] Figure 8 This is a flowchart showing the flow of the learning phase according to this embodiment.

[0018] Figure 9 This is a flowchart showing the flow of the measurement phase according to this embodiment. DETAILED DESCRIPTION

[0019] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, in this specification and the accompanying drawings, components having substantially the same functional structure are denoted by the same reference numerals to omit repeated description.

[0020] <Configuration example>

[0021] (Learning device 10)

[0022] The learning device 10 of this embodiment can be a device that performs supervised learning using the same sensor data acquired synchronously on a time axis using two different methods as input. Supervised learning refers to a method in which a pair of input data (learning data) and correct answer data (training data) corresponding to the input data is provided to a computer, causing the computer to learn the correspondence between the two. Figure 1 1 is a diagram showing an example of the functional configuration of the learning device 10 according to this embodiment. Figure 1 As shown, the learning device 10 according to this embodiment may include a learning unit 110 and a storage unit 120 .

[0023] One of the features of the learning unit 110 involved in this embodiment is that it uses first sensor data as learning data and training data based on second sensor data to learn outputs related to object feature points. The first sensor data is acquired using a first method and has a time length corresponding to a repetitive interval that can be periodically observed as time progresses. The second sensor data is acquired using a second method that is less affected by noise than the first method. The data is acquired at a predetermined time from the start of the time length related to the first sensor data, and the object feature point is the object observed in the repetitive interval. In addition, the predetermined time can be set based on the time length from the start of the repetitive interval to the time when the object feature point is expected to appear. According to this structure, it is possible to learn features such as data transitions after the object feature point, and to construct a more accurate learned model.

[0024] The learning unit 110 according to this embodiment can perform the above-described learning using any machine learning method that can implement supervised learning. The learning unit 110 performs learning using, for example, an algorithm such as a neural network or a support vector machine (SVM).

[0025] The functions of the learning unit 110 are realized by a processor such as a GPU (Graphics Processing Unit). The details of the functions of the learning unit 110 according to this embodiment will be described separately.

[0026] The storage unit 120 according to this embodiment stores various information related to the operation of the learning device 10. The storage unit 120 stores, for example, first sensor data and second sensor data used for learning by the learning unit 110, various parameters, and the like.

[0027] The functional configuration example of the learning device 10 according to this embodiment has been described above. Figure 1 The above-described structure is merely an example, and the structure of the learning device 10 according to this embodiment is not limited to this example. The learning device 10 according to this embodiment may further include, for example, an operation unit for receiving operations performed by an operator, an output unit for outputting various data, and the like. The structure of the learning device 10 according to this embodiment can be flexibly modified depending on specifications and applications.

[0028] Next, an example of the functional configuration of the measurement device 20 according to this embodiment will be described. The measurement device 20 according to this embodiment can be a device that uses the learned model constructed by the learning device 10 to perform measurements related to object feature points observed in sensor data acquired over time. Figure 2 1 is a diagram showing an example of the functional configuration of the measuring device 20 according to this embodiment. Figure 2 As shown, the measurement device 20 according to this embodiment may include an acquisition unit 210 and a measurement unit 220 .

[0029] The acquisition unit 210 according to this embodiment is configured to acquire first sensor data over time. Therefore, the acquisition unit 210 according to this embodiment includes various sensors corresponding to the characteristics of the acquired first sensor data.

[0030] The measurement unit 220 of this embodiment takes the first sensor data acquired by the acquisition unit 210 as input and measures the object feature points observed in the first center data. In this case, the measurement unit 220 of this embodiment uses a learned model constructed through learning by the learning unit 110 to produce outputs related to the object feature points. Specifically, one of the characteristics of the measurement unit 220 of this embodiment is that it measures the object feature points using a learned model obtained by learning the outputs related to the object feature points in a repetitive interval periodically observed over time using first sensor data having a time length corresponding to the repetitive interval as learning data and training data based on second sensor data acquired at a predetermined time after the start of the time length associated with the first sensor data.

[0031] According to the above configuration, it is possible to accurately measure the object feature points while effectively eliminating the influence of noise from the first sensor data. The functions of the measurement unit 220 according to this embodiment are realized by various processors.

[0032] The functional configuration example of the measuring device 20 according to this embodiment has been described above. Figure 2 The above-described configuration is merely an example, and the functional configuration of the measurement device 20 according to this embodiment is not limited to this example. The measurement device 20 according to this embodiment may further include an operation unit or output unit, an analysis unit for analyzing the measured object feature points, a notification unit for issuing various notifications based on the analysis results, and the like. The configuration of the measurement device 20 according to this embodiment can be flexibly modified depending on the characteristics of the object feature points being measured, their intended use, and the like.

[0033] Details

[0034] Next, the sensor data involved in this embodiment will be described using a specific example. In recent years, devices for acquiring various types of sensor data have been developed. Furthermore, even when acquiring the same type of sensor data, there are sometimes multiple methods. The sensor data described above includes vital data representing the subject's vital signs. Here, as an example of vital data, assume the case where the voltage changes caused by the subject's heart activity are acquired as an electrocardiogram waveform.

[0035] Methods for acquiring ECG waveforms include attaching multiple electrodes directly to the subject's skin and recording voltage changes through these electrodes, such as the three-point induction method and the standard 12-induction method. These methods can produce high-precision ECG waveforms with minimal noise influence. However, these methods often restrict the subject's behavior and, because the electrodes are attached directly to the skin, can be inconvenient for the subject.

[0036] In addition, as other methods of obtaining electrocardiographic waveforms, there is a method of setting electrodes at multiple locations that are expected to be in contact with the subject, and recording the changes in voltage obtained when the subject is in contact with multiple electrodes. Such a method is used, for example, in cases where it is desired to obtain the electrocardiographic waveform of a subject who is operating a device. As an example, it is known that there is a technology for obtaining the electrocardiogram of a driver by placing electrodes on a steering wheel or a driver's seat that is expected to be in contact with a driver of a moving object such as a vehicle. According to this technology, since it is not necessary to directly attach the electrodes to the driver's skin, the electrocardiographic waveform can be obtained without the driver being aware of it. On the other hand, in this case, noise is easily generated due to the body movement of the driver accompanying the driving behavior, the vibration of the vehicle, etc., and there is a possibility that the accuracy of the obtained electrocardiographic waveform will be reduced.

[0037] As described above, various methods for acquiring sensor data each have their own advantages, but there are also cases where the accuracy of the acquired sensor data varies. Therefore, a technology is needed that effectively utilizes the advantages of a particular method while simultaneously improving the accuracy of sensor data acquisition.

[0038] To address the above issues, the learning unit 110 of this embodiment uses first sensor data acquired using a first method as learning data and performs learning using training data based on second sensor data. The second sensor data is acquired synchronously with the first sensor data on the time axis using a second method that is less affected by noise than the first method. This effectively eliminates the effects of noise from the first sensor data and enables high-precision measurement of object feature points.

[0039] On the other hand, at this time, when the second sensor data corresponding to the end of the time length of the first sensor data or the second sensor data obtained after the end is used as training data, it is difficult for the learning unit 110 to learn information such as data transition after the training data.

[0040] In view of the above problems, the learning unit 110 involved in this embodiment can use as learning data first sensor data having a time length corresponding to a repetitive interval that can be periodically observed as time progresses. In addition, the learning unit 110 involved in this embodiment can use training data based on second sensor data obtained at a time when a specified time has passed since the start of the above time length. Here, the above specified time can be set based on the length of time from the start of the repetitive interval to the time when the object feature point is expected to appear. In this way, learning can be performed using information before and after the training data, thereby constructing a more accurate learned model.

[0041] Furthermore, the aforementioned repetitive interval may also include at least one other feature point that appears regularly on the time axis with the object feature point. In this case, by learning the temporal regularity between the object feature point and the other feature points in the repetitive interval, a learned model can be constructed that enables higher-precision measurement of the object feature point.

[0042] The following description uses as an example a case where the first and second sensor data involved in this embodiment are each an electrocardiogram waveform recording the heart activity of a subject. Specifically, the first sensor data involved in this embodiment can be a first electrocardiogram waveform obtained from the subject using a first method. Furthermore, the second sensor data can be a second electrocardiogram waveform obtained from the subject using a second method.

[0043] In addition, in this case, the above-mentioned first method can be a method of obtaining an ECG waveform using at least two electrodes expected to be in contact with the subject, and the above-mentioned second method can be a method of obtaining an ECG waveform using at least three electrodes directly worn on the subject's skin (for example, a three-point induction method).

[0044] For example, when the subject is a driver of a mobile object such as a vehicle, the two electrodes used in the first embodiment may be provided on a seat where the subject sits and an operated device (eg, a steering wheel) operated by the subject.

[0045] According to the above configuration, it is possible to obtain high-precision data that eliminates noise caused by the driver's body movements and vehicle vibrations while maintaining the advantage of the second embodiment, such as not causing inconvenience to the driver.

[0046] Here, characteristic points (characteristic waveforms) in a general electrocardiographic waveform will be described. Figure 3 is a diagram showing an example of a general electrocardiogram waveform in one cycle. Figure 3 In the figure, the horizontal axis represents the passage of time, and the vertical axis represents the change in voltage. Figure 3 As shown, a typical electrocardiogram waveform can be observed to have multiple characteristic waveforms representing characteristic shapes. Examples of characteristic waveforms include the P wave, Q wave, R wave, S wave, QRS wave (formed by the Q wave, R wave, and S wave), T wave, and U wave. Furthermore, each of these characteristic waveforms has a regularity of appearing in the order listed above on the time axis.

[0047] Among them, for example, the R wave is an important characteristic waveform as an indicator of heart rate variation (fluctuation). The interval between the R wave in a certain cycle and the R wave in the next cycle (RRI: RR Interval) is used to calculate the cycle of the heart rate. In addition, it is also known that RRI fluctuates due to stress or fatigue, and is also an effective physiological indicator when detecting the physical or psychological burden of the subject. In addition, for example, the interval between the Q wave and the T wave in one cycle, that is, QTI (QT Interval), represents the time from the start of ventricular excitement to the end of excitement, and is an important physiological indicator for the detection of arrhythmia.

[0048] Thus, one cycle of the electrocardiogram waveform contains multiple characteristic waveforms useful for obtaining physiological indices. Therefore, in the learning of this embodiment, the entire cycle can be set as a repetition interval, and the characteristic waveform corresponding to any physiological indices to be obtained can be used as the target feature point.

[0049] On the other hand, in one cycle of the electrocardiogram waveform, there is a section where characteristic waveforms useful for obtaining physiological indices are concentrated. Figure 3 As shown in FIG. 1 , the P wave, Q wave, R wave, S wave, and T wave can be observed continuously for a period of approximately 700 ms. Therefore, in the learning of this embodiment, the interval from the start of the P wave to the end of the T wave can be set as a repetition interval. This allows for more accurate learning of the temporal regularity between the P wave, Q wave, R wave, S wave, and T wave within the repetition interval.

[0050] Furthermore, for example, in the aforementioned repetitive interval, the R wave can be observed approximately 250 ms from the start of the P wave. Thus, using the R wave as the target feature point, the learning unit 110 of this embodiment learns the output related to the R wave using training data based on the second sensor data acquired from the start of the time length (700 ms) related to the first sensor data until the time length (250 ms) from the start of the P wave to the expected appearance of the R wave has elapsed.

[0051] Figure 4 This is a diagram showing an example of correspondence between learning data and training data according to this embodiment. Figure 4 The upper portion of represents the first sensor data (first cardiac waveform) obtained from the subject. Figure 4 The lower portion of represents second sensor data (second cardiac waveform) acquired from the subject during the same period as the first sensor data.

[0052] In this case, for example, the learning unit 110 can use the first sensor data obtained in the time length d1 corresponding to the 700 ms interval from the start time of the P wave to the end time of the T wave as the learning data of the first sequence, and also use the second sensor data obtained at the time t1 when 250 ms has passed from the start time of the time length d1 as the training data for learning.

[0053] Similarly, the learning unit 110 can perform learning by using the first sensor data obtained in the time length d2 as the learning data of the second sequence, and the second sensor data obtained at time t2 when 250 ms have passed from the start time of the time length d2 as the training data.

[0054] Similarly, the learning unit 110 can perform learning by using the first sensor data obtained in the time length d3 as the learning data of the third sequence, and the second sensor data obtained at time t3 when 250ms have passed since the start time of the time length d3 as the training data.

[0055] Using the aforementioned dataset, the temporal regularity between the R wave and other characteristic waveforms included in the repetitive period can be effectively learned. Furthermore, the measurement unit 220 of this embodiment can accurately measure R waves using a learned model constructed through learning using the aforementioned dataset. Figure 5 1 and 2 are diagrams showing an image of object feature points measured by the measurement unit 220 according to the present embodiment.

[0056] like Figure 5As shown, the measuring unit 220 of this embodiment is provided by Figure 4 The learned model constructed using the illustrated dataset takes as input the first sensor data (first cardiac waveform) and outputs third sensor data (third cardiac waveform) obtained by removing noise from the first sensor data. This allows for highly accurate R-wave measurement based on the third sensor data, even when direct R-wave measurement from the first sensor data is difficult due to noise.

[0057] In addition, the above description describes a case where the learning unit 110 uses the second sensor data itself (for example, the voltage value of the second cardiac waveform) as training data for learning, but the learning unit 110 involved in this embodiment can also use the existence probability data representing the existence probability of the object feature point in the second sensor data as training data to learn the output related to the existence probability of the object feature point in the first sensor data.

[0058] For example, in Figure 4 In the example shown, the learning unit 110 performs learning on the first sensor data (learning data) acquired during time length d1, using the R-wave presence probability data generated based on the second sensor data acquired at time t1 as training data. When the presence probability data represents the probability of the R-wave presence using either 0 (absence) or 1 (presence), since the R-wave exists at time t1, the R-wave presence probability data at time t1 is 1. On the other hand, the R-wave does not exist at time t2 or time t3. Therefore, the R-wave presence probability data at time t2 and time t3 is 0.

[0059] When learning is performed using the above-mentioned existence probability data as training data, as shown in FIG. Figure 6 As shown, the measurement unit 220 of this embodiment can directly output R-wave presence probability data by inputting the first sensor data (first cardiac waveform) into the learned model. This allows the learning of this embodiment to use training data corresponding to the desired data format to be output by the measurement unit 220. Furthermore, while the above example illustrates the case where the presence probability data takes two values, 0 or 1, the presence probability data of this embodiment can also take three or more values.

[0060] Here, the results of verification of the R-wave detection accuracy using the learned model constructed through learning according to this embodiment are shown. Figure 7 : is a graph showing the detection accuracy of the R wave using the learned model involved in this embodiment. Figure 7This represents the R-wave detection accuracy using the learned model constructed when the time length of the learning data (first sensor data) was set to 500ms, 600ms, 700ms, and 800ms. In all cases, learning was performed using training data based on the second sensor data acquired 250ms after the start of the time length of the learning data.

[0061] The results, such as Figure 7 As shown, the learned model constructed using 700ms of learning data can detect R waves with the highest accuracy. This verification result shows that more efficient learning can be achieved by setting the time length of the learning data in accordance with the regularity of the object feature points and other feature points on the time axis.

[0062] On the other hand, the setting of a time length of 700ms is just an example. It is assumed that the optimal time length of the learning data varies based on the statistical characteristics of the first sensor data used as the learning data. For example, in the first sensor data obtained under a certain condition, when the average time length from the start time of the P wave to the end time of the T wave is 650ms, the time length of the learning data can also be set to 650ms. In addition, the time length of the training data is the same. For example, in the first sensor data and the second sensor data obtained, when the average time length from the start time of the P wave to the R wave is 300ms, training data based on the second sensor data obtained at a time when 300ms has passed since the start time of the time length involved in the learning data can also be used.

[0063] <Flow of the Learning and Measurement Phases>

[0064] Next, the flow of a learning phase in which learning is performed using the learning device 10 according to the present embodiment and a measurement phase in which measurement is performed using the measurement device 20 will be described. Figure 8 This is a flowchart showing the flow of the learning phase according to this embodiment.

[0065] like Figure 8 As shown, in the learning phase of this embodiment, first, first sensor data and second sensor data are acquired (S102). At this time, the first sensor data and second sensor data may be acquired along with information such as a timestamp to synchronize them on the time axis. Alternatively, the first sensor data and second sensor data may be acquired by a device separate from the learning device 10. The acquired first sensor data and second sensor data are stored in the storage unit 120 of the learning device 10.

[0066] Next, the first and second sensor data are processed as needed (S104). For example, when using presence probability data related to object feature points as training data, step S104 may convert the second sensor data obtained in step S102 into presence probability data. Alternatively, various filtering processes may be performed to reduce noise contained in the first and second sensor data. Furthermore, the aforementioned processing may be performed by a device separate from the learning device 10.

[0067] Next, the learning unit 110 performs learning using the first sensor data having a time length corresponding to the repetition interval as learning data and the training data based on the second sensor data acquired at a time point after a predetermined time has elapsed from the start time of the repetition interval (S106). In this case, the learning unit 110 may use the second sensor data itself (or the filtered second sensor data) as training data, or may use the presence probability data generated in step S104 as training data.

[0068] The above describes the flow of the learning phase according to the present embodiment. Next, the flow of the measurement phase according to the present embodiment will be described. Figure 9 This is a flowchart showing the flow of the measurement phase according to this embodiment.

[0069] like Figure 9 As shown, in the measurement phase according to this embodiment, the acquisition unit 210 first acquires first sensor data using a first method (S202). The acquisition unit 210 may acquire the driver's electrocardiogram waveform as the first sensor data, for example, using a plurality of electrodes disposed on the steering wheel and seat of the vehicle.

[0070] Next, the measurement unit 220 inputs the first sensor data acquired in step S202 into the learned model and measures the object feature points contained in the first sensor data (S204). If the second sensor data was used as training data during the learning phase, the measurement unit 220 outputs third sensor data obtained by removing noise from the first sensor data and measures the object feature points. On the other hand, if the presence probability data was used as training data during the learning phase, the measurement unit 220 outputs presence probability data indicating the presence probability of the object feature points and measures the object feature points.

[0071] Next, various actions based on the target feature points measured in step S204 are performed as needed (S206). For example, if the target feature points are R waves, the actions may include notification based on the RRI. These actions may also be performed by a device other than the measurement device 20.

[0072] <Supplement>

[0073] While the preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, the present invention is not limited to these embodiments. Those skilled in the art will be able to devise various variations or modifications within the scope of the technical concept described in the claims, and these should naturally be understood to fall within the technical scope of the present invention.

[0074] For example, in the above embodiment, the example in which the learning unit 110 learns measurements related to the subject's cardiac activity is mainly described. However, the learning target of the learning unit 110 is not limited to the aforementioned vital data measurements. For example, the learning unit 110 can also measure various data indicating the operating status of any device.

[0075] In addition, in the above embodiment, as a first method for obtaining an electrocardiographic waveform, a method of disposing electrodes at a location expected to be in contact with the subject is exemplified, and as a second method, a method of directly wearing the electrodes on the subject's skin is exemplified. On the other hand, the first and second methods in the present technology may be any different methods that differ in the degree to which they are affected by noise. For example, in the case of obtaining heart rate, the first method may be a non-contact method using a Doppler sensor. In this case, the second method may be any method in which the influence of noise is less than that of the non-contact method. For example, the second method in the above case may be a contact method in which electrodes are worn on the subject's skin. Thus, the first method in the present technology is not limited to the methods exemplified in the above embodiment and may be selected as appropriate. Furthermore, in the case where a contact method for obtaining an electrocardiographic waveform using at least two electrodes expected to be in contact with the subject is less affected by noise than a non-contact method using a Doppler sensor, etc., the non-contact method may be set as the first method and the contact method may be set as the second method.

[0076] In addition, the series of processes performed by each device described in this invention can also be implemented using any one of software, hardware, and a combination of software and hardware. The programs constituting the software are, for example, pre-stored in a recording medium (non-transitory media) provided inside or outside each device. Moreover, each program is read into RAM, for example, when executed by a computer, and executed by a processor such as a CPU. The above-mentioned recording medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, etc. In addition, the above-mentioned computer program can also be distributed, for example, via a network, without using a recording medium.

[0077] Description of Reference Numerals

[0078] 10 ...learning device; 110 ...learning unit; 120 ...storage unit; 20 ...measuring device; 210 ...acquisition unit; 220 ...measuring unit.

Claims

1. A learning device, characterized in that: The learning device includes a learning unit that uses the first sensor data as learning data and training data based on the second sensor data to learn outputs related to feature points of the object. The first sensor data is obtained by a first method and has a time length corresponding to a repetitive interval that can be periodically observed as time progresses. The second sensor data is data acquired at a time when a predetermined time has elapsed from the start time of the time length involved in the first sensor data using a second method in which the influence of noise is less than that of the first method. The object feature point is an object observed in the repetition interval, The predetermined time is set based on the length of time from the start time of the repetition section to the time when the object feature point is expected to appear.

2. The learning device according to claim 1, wherein The repetitive section includes at least one other feature point that has a regularity in appearance with the object feature point on the time axis.

3. The learning device according to claim 1 or 2, characterized in that The first sensor data and the second sensor data are electrocardiographic waveforms recording the heart activity of the subject.

4. The learning device according to claim 3, wherein The repetitive section is a section from the start time of the P wave to the end time of the T wave.

5. The learning device according to claim 4, characterized in that The object feature point is the R wave, The learning unit uses training data based on the second sensor data to learn the output related to the R wave, and the second sensor data is data obtained at the moment of the start of the time length involved in the first sensor data, and the time length from the start of the P wave to the moment when the R wave is expected to appear.

6. The learning device according to claim 5, characterized in that The learning unit learns an output related to the existence probability of the R wave in the first sensor data using existence probability data indicating the existence probability of the R wave in the second sensor data as training data.

7. The learning device according to claim 3, wherein: The first method is a method of obtaining an electrocardiogram waveform using at least two electrodes that are expected to be in contact with the subject. The second method is a method of obtaining an electrocardiographic waveform using at least three electrodes worn on the skin of the subject.

8. The learning device according to claim 3, wherein: The test subject is a driver of a mobile vehicle.

9. A learning method, characterized in that include: First sensor data is used as learning data, the first sensor data being acquired by a first method and having a time length corresponding to a repetitive interval that can be periodically observed as time progresses, Using training data based on second sensor data, the second sensor data is data acquired at a time point after a predetermined time has elapsed from the start time of the time length involved in the first sensor data, using a second method in which the influence of noise is less than that of the first method; learning the outputs related to the object feature points observed in the repetition interval, The predetermined time is set based on a time length from a start time of the repetition section to a time when the object feature point is expected to appear.

10. A measuring device, characterized in that: The measuring device includes a measuring unit that receives first sensor data acquired by a first method and performs measurement on an object feature point to be observed in the first sensor data. The measurement unit performs measurements related to the object feature points using a learned model, the learned model being obtained by learning outputs related to the object feature points in a repetitive interval using the first sensor data having a time length corresponding to a repetitive interval that can be periodically observed as time progresses, and using training data based on second sensor data, wherein the second sensor data is obtained at a time point after a predetermined time has passed from a start time point of the time length related to the first sensor data using a second method in which the influence of noise is smaller than that of the first method. The predetermined time is set based on the length of time from the start time of the repetition section to the time when the object feature point is expected to appear.

Images