Identification method, identification device, control program, and recording medium

Abstract

The present invention accurately identifies an event observed as time-series data. This identification method comprises: an acquisition step (S101) for acquiring first time-series data obtained by measuring a first event and second time-series data obtained by measuring a second event; a decomposition step (S102) for obtaining first mode information obtained by decomposing the first time-series data into a plurality of dynamic modes and second mode information obtained by decomposing the second time-series data into a plurality of dynamic modes; and an identification step (S104) for identifying the first event and the second event on the basis of the result of comparison between the first mode information and the second mode information.

Description

Identification method, identification device, control program, and recording medium

[0001] The present invention relates to a method and an apparatus for identifying events observable as time-series data.

[0002] In recent years, the usefulness of applying dynamic mode decomposition (DMD) to multidimensional time series analysis has been widely studied. For example, Non-Patent Document 1 proposes a technique for detecting epileptic seizures based on features extracted by wavelet analysis of surface EEGs and features obtained by DMD of the surface EEGs.

[0003] Deba Prasad Dash, Maheshkumar H. Kolekar, Kamlesh Jha, “Surface EEG based epileptic seizure detection using wavelet based features and dynamic mode decomposition power along with KNN classifier”, Multimedia Tools and Applications, 81:42057-42077, 2022.

[0004] A classification method and a classification device are provided that accurately classify events observed as time-series data.

[0005] An object of one aspect of the present invention is to provide an identification method, an identification device, and the like that can be applied to events that can be observed as time-series data.

[0006] In order to solve the above-mentioned problems, an identification method according to one aspect of the present invention is an identification method executed by one or more computers, and includes: an acquisition step of acquiring first time series data measuring a first event and second time series data measuring a second event; a decomposition step of decomposing each of the first time series data and the second time series data into a plurality of dynamic modes by dynamic mode decomposition, and calculating first mode information regarding the plurality of dynamic modes of the first time series data and second mode information regarding the plurality of dynamic modes of the second time series data; and an identification step of identifying the first event and the second event based on a comparison result between the first mode information and the second mode information.

[0007] In order to solve the above problem, an identification device according to one aspect of the present invention includes: an acquisition unit that acquires first time series data measuring a first event and second time series data measuring a second event; a decomposition unit that decomposes each of the first time series data and the second time series data into a plurality of dynamic modes using dynamic mode decomposition to calculate first mode information regarding the plurality of dynamic modes of the first time series data and second mode information regarding the plurality of dynamic modes of the second time series data; and an identification unit that identifies the first event and the second event based on a comparison result between the first mode information and the second mode information.

[0008] The identification device according to each aspect of the present invention may be realized by a computer. In this case, the control program for the identification device that causes the computer to operate as each part (software element) of the identification device to realize the identification device, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention.

[0009] According to one aspect of the present invention, it is possible to realize a classification method, a classification device, and the like that can accurately classify events observed as time-series data.

[0010] 1 is a block diagram showing an example of the configuration of a classification system according to an aspect of the present invention; 2 is a diagram for explaining dynamic mode decomposition; 3 is a flowchart showing an example of the flow of processing performed by a classification device; 4 is a diagram showing an example of a result of classifying brain activity of a healthy subject and brain activity of an epilepsy patient; 5 is a diagram showing an example of classification accuracy when EEG data of a healthy subject and EEG data of an epilepsy patient are classified based on mode information; and 6 is a diagram showing an example of a result of classifying brain activity of a healthy subject and brain activity of an Alzheimer's patient.

[0011] [Embodiment 1] The inventors of the present invention have discovered that mode information obtained by decomposing time-series data obtained by measuring an event into multiple modes contains information that can be used as a feature that accurately represents the characteristics of the event. Therefore, the inventors of the present invention have arrived at the present invention, which uses mode information corresponding to each of multiple events to identify each of multiple events. One embodiment of the present invention will be described in detail below.

[0012] (Configuration of Identification System 100) The configuration of the identification system 100 including an identification device 1 capable of identifying multiple events observed as time-series data will be described with reference to FIG. 1. FIG. 1 is a block diagram showing an example configuration of the identification system 100 according to one aspect of the present invention. The identification system 100 includes an identification device 1 that acquires and analyzes measurement data measured by a measurement device 2. The measurement device 2 may be a device that measures spatial and temporal changes in each of multiple events as time-series data at each of multiple measurement points. While FIG. 1 shows a configuration in which one identification device 1 and multiple measurement devices 2 are communicatively connected, the identification system 100 is not limited to this configuration. For example, the identification system 100 may also be configured to include one measurement device 2. Note that in the identification system 100, the functions of the identification device 1 may be implemented by one or more computers.

[0013] (Events that can be identified by the identification device 1) The identification device 1 is a device that can identify multiple events observed as time-series data. There are no particular restrictions on the events that can be identified, and they are diverse. For example, an event that can be identified by the identification device 1 is brain activity of a living organism observed as time-series data. In this case, the measurement device 2 may be a device that measures data indicating brain activity of a living organism, and the identification system 100 may be configured to identify brain activity of a living organism of interest (first event) and brain activity of a control living organism (second event).

[0014] The living organism of interest is a living organism having some abnormality in its brain activity, and the control living organism may be a living organism having normal brain activity or a living organism suffering from a specific disease. Alternatively, the living organism of interest may be a sleeping living organism, and the control living organism may be a waking living organism. The identification system 100 can accurately distinguish between the brain activity of the living organism of interest and the brain activity of the control living organism.

[0015] The living organism of interest may be a living organism performing a predetermined task A, and the control living organism may be a living organism performing a predetermined task B. Task A and task B may be the same type of task, or different types of tasks. If tasks A and B are the same type of task, the identification system 100 can output an identification result indicating that the brain activity of the living organism of interest and the brain activity of the control living organism are the same. On the other hand, if tasks A and B are different types of tasks, the identification system 100 can output an identification result indicating that the brain activity of the living organism of interest and the brain activity of the control living organism are different.

[0016] The organism of interest may be an organism suffering from a certain disease A, and the control organism may be an organism suffering from a certain disease B. Disease A and disease B may be the same disease, or they may be different diseases. If disease A and disease B are the same disease, the identification system 100 can output an identification result indicating that the brain activity of the organism of interest and the brain activity of the control organism are the same. On the other hand, if disease A and disease B are different diseases, the identification system 100 can output an identification result indicating that the brain activity of the organism of interest and the brain activity of the control organism are different.

[0017] In this way, the identification system 100 may identify the brain activity of the living organism of interest and the brain activity of the control living organism. In this case, the data measuring the brain activity of the living organism of interest (first time-series data) may be electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the living organism of interest. Furthermore, the data measuring the brain activity of the control living organism (second time-series data) may be electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the control living organism.

[0018] Alternatively, the identification system 100 may identify a displacement pattern of multiple parts of a moving target object (first event) and a displacement pattern of multiple parts of a moving control object (second event). In this case, the data measuring the displacement pattern of multiple parts of the moving target object may be time-series data measuring the displacement of each of the multiple parts of the target object over time (first time-series data). Furthermore, the data measuring the displacement pattern of multiple parts of a moving control object may be time-series data measuring the displacement of each of the multiple parts of the control object over time (second time-series data).

[0019] The target object and the control object may be living organisms that move or change their posture through movement. If the target object and the control object are human, the multiple parts may be any parts that move with movement, such as the head, neck, arms, legs, or joints. Alternatively, the target object and the control object may be components of inanimate objects, such as various devices and robots, that move or change their posture through movement. With this configuration, the identification system 100 can output an identification result indicating whether the movement of the test object component differs from the movement of a normal component, for example, based on a comparison result between time-series data (first time-series data) indicating the movement of the test object component over time and time-series data (second time-series data) indicating the movement of a normally operating component over time.

[0020] Alternatively, the target object and the control object may be a collection of living and / or inanimate objects that move over time in a certain location. For example, data measuring the displacement pattern of multiple parts of a moving target object may be data measuring the movement of people and vehicles passing through a certain passage A (first time-series data). Furthermore, data measuring the displacement pattern of multiple parts of a moving control object may be data measuring the movement of people and vehicles passing through a certain passage B (second time-series data).

[0021] The identification system 100 may identify a pattern of increase or decrease in the number of cells contained in an object of interest (first event) and a pattern of increase or decrease in the number of cells contained in a control object (second event). In this case, the object of interest and the control object may be a collection of cells at a certain location or under certain conditions (e.g., a colony of cultured cells and biological tissue). For example, the data measuring the pattern of increase or decrease in the number of cells contained in the object of interest may be data (first time-series data) showing a pattern of change over time in the number of cells measured from image data of a colony of cells cultured under certain culture condition A. Furthermore, the data measuring the pattern of increase or decrease in the number of cells contained in the control object may be data (second time-series data) showing a pattern of change over time in the number of cells measured from image data of a colony of cultured cells cultured under certain culture condition B.

[0022] The classification device 1 is not limited to a configuration that classifies multiple events, and may be configured to perform regression analysis to estimate a predetermined value related to the events. Furthermore, the classification device 1 may be configured to perform classification using a machine learning model that receives time-series data or mode information as input and outputs a classification result or the predetermined value. In the aforementioned configuration that performs machine learning, the storage unit 20 stores, for example, a parameter set that defines a neural network used for machine learning, and the control unit 10 performs training of the machine learning model by updating the values ​​of the parameter set.

[0023] (Configuration of Identification Apparatus 1) Next, the configuration of the identification apparatus 1 will be described with reference to Fig. 1. As shown in Fig. 1, the identification apparatus 1 includes a control unit 10 and a storage unit 20.

[0024] The control unit 10 can be realized by a logic circuit (hardware) formed on an integrated circuit (IC chip) or the like, or can be realized by software. When realized by software, the control unit 10 may be configured, for example, by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a combination thereof. As shown in FIG. 1 , the control unit 10 may include an acquisition unit 11, a decomposition unit 12, an identification unit 13, and an output unit 14.

[0025] The storage unit 20 is a storage device that at least temporarily stores software and various data used by the identification device 1. As the storage unit 20, for example, a random access memory (RAM), a hard disk drive (HDD), a solid-state drive (SSD), a secure digital (SD) card, or an embedded multi-media controller (eMMC) can be used.

[0026] The acquisition unit 11 acquires first time series data measuring a first event and second time series data measuring a second event. The identification device 1 may be configured to acquire the first time series data and the second time series data from the measurement device 2, as shown in FIG. 1 . However, the identification device 1 is not limited to this configuration. For example, the identification device 1 may be configured such that the first time series data and the second time series data are stored in advance in the storage unit 20, and the identification device 1 reads (acquires) this time series data from the storage unit 20 and executes processing. Alternatively, the identification device 1 may be configured to acquire the first time series data and the second time series data input by a user as data to be processed and execute processing.

[0027] The decomposition unit 12 decomposes each of the first time series data and the second time series data into a plurality of dynamic modes. The decomposition unit 12 may apply any decomposition method capable of decomposing each of the first time series data and the second time series data into a plurality of dynamic modes. The decomposition unit 12 calculates first mode information regarding the plurality of modes for the first time series data and second mode information regarding the plurality of dynamic modes for the second time series data. Dynamic mode decomposition that can be applied by the decomposition unit 12 will be described below.

[0028] Dynamic mode decomposition is a method for decomposing high-dimensional time series data into the frequencies of multiple spatial features and time-domain features corresponding to the spatial features. When the time series data contains a phenomenon with multiple frequencies, the dynamic mode represents the dynamic mode eigenvalues ​​having each frequency and the spatial feature points operating at each frequency (i.e., the positions of the phenomenon at each frequency).

[0029] Dynamic mode decomposition is an operation for reducing the dimension of high-dimensional time series data to obtain mode information including frequency information corresponding to each dynamic mode. By dynamic mode decomposition, X(t), which represents each of the first time series data and the second time series data, is approximated by the following Equation 1:

[0030] Here, Φ on the right-hand side is an eigenvalue vector corresponding to each of the multiple dynamic modes. Also, exp(Ωt) on the right-hand side is an eigenvalue, and Ω has a complex value, with the imaginary part of the value corresponding to the frequency and the real part corresponding to the increase or decrease of each frequency component. t on the right-hand side corresponds to the measurement time of the time series data. z on the right-hand side is a value corresponding to the initial value (amplitude) of the dynamic mode corresponding to each of the first time series data and the second time series data at time 0.

[0031] <About Dynamic Mode Decomposition> Time series data of a certain event measured at multiple measurement points by the measurement device 2 can be approximated as being made up of multiple overlapping waveforms with different rates of change (i.e., attenuation or increase) in frequency and amplitude. Here, the dynamic mode decomposition performed by the decomposition unit 12 will be explained using simplified time series data as an example with reference to Figure 2. Figure 2 is a diagram for explaining dynamic mode decomposition.

[0032] The graph 30 shown in FIG. 2 shows a waveform X whose amplitude peaks at a certain position (Position(x) is negative). １ (x, t) and a waveform X with a peak amplitude at another position (Position(x) is positive). ２ (x, t) is added to the waveform X(x, t). １ (x, t) is a waveform with a frequency of 13 Hz and an amplitude that attenuates by 0.25 times per unit time, and waveform X ２ (x, t) is a waveform with a frequency of 8 Hz and an amplitude that doubles per unit time. That is, the waveform shown in graph 30 is a waveform obtained by superimposing the waveform shown in graph 31a and the waveform shown in graph 31c.

[0033] The coordinate axis "Position(x)" on the lower right side of the graph 30 corresponds to, for example, the position of the measurement point, and the coordinate axis "Time(t)" on the lower left side corresponds to time. Note that it is not necessary to make the coordinate axis "Position(x)" correspond to the position of the measurement point. For example, changing the order in which the coordinate axis "Position(x)" is plotted does not affect the identification results. Furthermore, the positions of the measurement points may be represented on two-dimensional or three-dimensional coordinates, or may be represented by positions mapped to the measurement object. For example, if the population trends of each country are being observed, the position of each country on a map may be associated with each measurement point.

[0034] The vertical axis of graph 30 is a coordinate axis corresponding to the measured value. In this way, time-series data measuring a certain event can be displayed on a three-dimensional coordinate system defined by an axis indicating the position of the measurement point ("Position(x)"), an axis indicating time ("Time(t)"), and an axis indicating the magnitude of the measured value. Note that in graphs 31a, 31c, 36a-36d, 37a-37d, 38, and 39, the axes "Position(x)" and "Time(t)" may be omitted.

[0035] When dynamic mode decomposition is applied to the waveform X(x, t) shown in graph 30, it is decomposed into DMD component 1 shown in graph 32a, DMD component 2 shown in graph 32b, DMD component 3 shown in graph 32c, and DMD component 4 shown in graph 32d. Here, DMD components 1 and 2 are decomposed into the waveform X(x, t) shown in graph 31a. １ The mode (Mode(φ)) corresponding to (x, t) is the waveform X shown in graph 31b. ２ The mode (Mode(φ)) corresponding to (x, t) is shown in graph 35. Furthermore, as shown in data 33 and 34, DMD components 1 to 4 are decomposed into time dynamics in the real part of the measured values ​​and time dynamics in the imaginary part of the measured values. That is, by applying dynamic mode decomposition to waveform X(x, t) shown in graph 30, it is possible to obtain the mode (Mode(φ)) corresponding to each of the multiple waveforms constituting waveform X(x, t) and the time dynamics of the measured values ​​corresponding to each of the multiple waveforms. The mode (Mode(φ)) corresponding to each of the multiple waveforms and the time dynamics of the measured values ​​corresponding to each of the multiple waveforms are mode information for waveform X(x, t), which is time series data. The mode information for the time series data includes the time dynamics of the measured values ​​corresponding to each of the multiple waveforms, i.e., frequency information related to the frequencies corresponding to each of the multiple dynamic modes of the time series data.

[0036] If graphs 33, 34, and 35 are plotted on a three-dimensional coordinate system defined by an axis indicating the position of the measurement point (Position(x)), an axis indicating time (Time(t)), and an axis indicating the magnitude of the measurement value, graphs 36 and 37 are obtained. Here, graph 36 shows a three-dimensional plot of the real parts of the measurement values, and graph 37 shows a three-dimensional plot of the imaginary parts of the measurement values. Graphs 36a and 37a, 36b and 37b, 36c and 37c, and 36d and 37d correspond to DMD components 1 to 4, respectively. Adding the real parts of the measurement values ​​shown in graphs 36a to 36d results in a waveform Xrecon(x,t) similar to the waveform X(x,t) shown in graph 30, as shown in graph 38. On the other hand, adding the imaginary parts of the measurement values ​​shown in graphs 37a to 37d cancels out all the measurement values.

[0037] In this way, the decomposition unit 12 can calculate, from each of the first time series data and the second time series data, first mode information relating to a plurality of modes of the first time series data and second mode information relating to a plurality of dynamic modes of the second time series data. The first mode information and the second mode information accurately reflect the characteristics of the first time series data and the second time series data, respectively, and can therefore be used to distinguish between the first event and the second event.

[0038] Returning to Figure 1, the identification unit 13 identifies the first event and the second event based on the comparison result between the first mode information calculated for the first time series data and the second mode information calculated for the second time series data.

[0039] The first mode information may include first frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the first time series data, and the second mode information may include second frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the second time series data. In this case, the identification unit 13 may identify the first event and the second event based on a comparison result between the first frequency information and the second frequency information.

[0040] Here, the first frequency information may be a statistic indicating a frequency distribution of frequencies corresponding to each of a plurality of dynamic modes of the first time series data. Furthermore, the second frequency information may be a statistic indicating a frequency distribution of frequencies corresponding to each of a plurality of dynamic modes of the second time series data. For example, the first frequency information may be a number of numerical values ​​indicating frequencies corresponding to each of a plurality of dynamic modes of the first time series data that belong to a predetermined frequency band. Furthermore, the second frequency information may be a number of numerical values ​​indicating frequencies corresponding to each of a plurality of dynamic modes of the second time series data that belong to a predetermined frequency band.

[0041] Alternatively, the first frequency information may be a plurality of numerical values ​​indicating frequencies corresponding to the plurality of dynamic modes of the first time series data, and the second frequency information may be a plurality of numerical values ​​indicating frequencies corresponding to the plurality of dynamic modes of the second time series data.

[0042] The mode information may be expressed as a matrix in which frequency components corresponding to each dynamic mode are arranged in a matrix, and which indicates the increase or decrease of each frequency component.

[0043] The first mode information may include, for the first time series data, at least one of diagonal elements and off-diagonal elements of a feature matrix R1 obtained based on a matrix P1 in which elements corresponding to each of a plurality of dynamic modes are arranged in a matrix and a matrix Q1 that is a transpose of matrix P1.Furthermore, the second mode information may include, for the second time series data, at least one of diagonal elements and off-diagonal elements of a feature matrix R2 obtained based on a matrix P2 in which elements corresponding to each of a plurality of dynamic modes are arranged in a matrix and a matrix Q2 that is a transpose of matrix P2.

[0044] (Processing Performed by Identification Apparatus 1) Next, the flow of processing performed by the identification apparatus 1 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing an example of the flow of processing performed by the identification apparatus 1.

[0045] First, in step S101 (acquisition step), the acquisition unit 11 acquires, for example, from the measurement device 2, first time series data including time series data corresponding to a first event and second time series data including time series data corresponding to a second event.

[0046] Subsequently, in step S102 (decomposition step), the decomposition unit 12 decomposes each of the first time series data and the second time series data into a plurality of dynamic modes by dynamic mode decomposition.

[0047] Next, in step S103, the decomposing unit 12 calculates first mode information relating to a plurality of dynamic modes of the first time series data and second mode information relating to a plurality of dynamic modes of the second time series data.

[0048] Then, in step S104 (identification step), the identification unit identifies the first event and the second event based on the comparison result between the first mode information calculated for the first time series data and the second mode information calculated for the second time series data.

[0049] In step S105, the output unit 14 outputs the classification result to an output device (not shown). The output device may be, for example, a display device equipped with a display, a computer, or the like.

[0050] [Embodiment 2] The functions of the identification device 1 (hereinafter referred to as the "device") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control unit 10).

[0051] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using the control device and storage device.

[0052] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0053] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0054] Furthermore, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI ​​may run on the control device or on another device (for example, an edge computer or a cloud server).

[0055] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0056] [Summary] An identification method according to aspect 1 of the present invention is an identification method executed by one or more computers, and includes: an acquisition step of acquiring first time series data measuring a first event and second time series data measuring a second event; a decomposition step of obtaining first mode information by decomposing the first time series data into a plurality of dynamic modes and second mode information by decomposing the second time series data into the plurality of dynamic modes; and an identification step of identifying the first event and the second event based on a comparison result between the first mode information and the second mode information.

[0057] An identification method according to Aspect 2 of the present invention may be configured in the above-mentioned Aspect 1, wherein the first mode information includes first frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the first time series data, and the second mode information includes second frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the second time series data, and in the identification step, the first event and the second event are identified based on a comparison result between the first frequency information and the second frequency information.

[0058] In the identification method according to Aspect 3 of the present invention, in Aspect 2, the first frequency information may be a statistic indicating a frequency distribution of frequencies corresponding to each of the plurality of dynamic modes of the first time series data, and the second frequency information may be a statistic indicating a frequency distribution of frequencies corresponding to each of the plurality of dynamic modes of the second time series data.

[0059] In the identification method according to Aspect 4 of the present invention, in Aspect 2 described above, the first frequency information may be a number of numerical values ​​indicating frequencies corresponding to each of the plurality of dynamic modes of the first time series data belonging to a predetermined frequency band, and the second frequency information may be a number of numerical values ​​indicating frequencies corresponding to each of the plurality of dynamic modes of the second time series data belonging to the predetermined frequency band.

[0060] In the identification method according to Aspect 5 of the present invention, in Aspect 2 above, the first frequency information may be a numerical value indicating a frequency corresponding to each of the plurality of dynamic modes of the first time series data, and the second frequency information may be a numerical value indicating a frequency corresponding to each of the plurality of dynamic modes of the second time series data.

[0061] A classification method according to a sixth aspect of the present invention is related to any one of the first to fifth aspects above, wherein the first mode information may include, for the first time series data, at least one of diagonal components of a feature matrix R1 obtained based on a matrix P1 in which components corresponding to each of the plurality of dynamic modes are arranged in a matrix form and a matrix Q1 that is a transpose of the matrix P1, and off-diagonal components of the feature matrix R1; and the second mode information may include, for the second time series data, at least one of diagonal components of a feature matrix R2 obtained based on a matrix P2 in which components corresponding to each of the plurality of dynamic modes are arranged in a matrix form and a matrix Q2 that is a transpose of the matrix P2, and off-diagonal components of the feature matrix R2.

[0062] A discrimination method according to aspect 7 of the present invention is such that, in any of aspects 1 to 6 above, the first event and the second event are brain activities of a living organism of interest and a control organism, respectively, the first time series data may be electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the living organism of interest, and the second time series data may be electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the control organism.

[0063] An identification method according to aspect 8 of the present invention may be such that, in any of aspects 1 to 6 above, the first event and the second event are displacement patterns of multiple parts of a moving target object and a moving control object, respectively, the first time series data includes data measuring the displacement over time of each of the multiple parts of the target object, and the second time series data includes data measuring the displacement over time of each of the multiple parts of the control object.

[0064] An identification device according to a ninth aspect of the present invention includes an acquisition unit that acquires first time series data measuring a first event and second time series data measuring a second event; a decomposition unit that acquires first mode information obtained by decomposing the first time series data into a plurality of dynamic modes and second mode information obtained by decomposing the second time series data into the plurality of dynamic modes; and an identification unit that identifies the first event and the second event based on a comparison result between the first mode information and the second mode information.

[0065] A control program according to aspect 10 of the present invention is a control program for causing a computer to function as the identification device described in aspect 9 above, and is a control program for causing a computer to function as the acquisition unit, the decomposition unit, and the identification unit.

[0066] A recording medium according to an eleventh aspect of the present invention is a computer-readable recording medium on which the control program according to the tenth aspect is recorded.

[0067] The present invention will be further explained below with reference to examples shown in FIGS. 4 to 6, but the present invention is not limited to these.

[0068] 4 is a diagram showing an example of the results of distinguishing between brain activity of a healthy subject and brain activity of an epileptic patient. Here, the brain activity of the healthy subject is an example of a first event, and the brain activity of the epileptic patient is an example of a second event. The EEG data of the healthy subject is an example of first time-series data, and the EEG data of the epileptic patient is an example of second time-series data. The EEG data may also be potentials estimated using a current source measurement method based on signals measured by a magnetoencephalograph, which is a measurement device.

[0069] In Fig. 4, graph 51 shows the frequency distribution of frequencies corresponding to each of multiple dynamic modes of EEG data obtained from current values ​​estimated at 160 vertices on the brain surface of a healthy subject. On the other hand, graph 52 shows the frequency distribution of frequencies corresponding to each of multiple dynamic modes of EEG data obtained from current values ​​estimated at 160 vertices on the brain surface of an epileptic patient. Note that Fig. 4 also shows the average frequency and 95% confidence interval.

[0070] When the frequency distribution of frequencies corresponding to each of the multiple dynamic modes of EEG data of a healthy subject was compared with the frequency distribution of frequencies corresponding to each of the multiple dynamic modes of EEG data of an epileptic patient, significant differences were observed, for example, in the range of 40 to 80 Hz and 100 to 160 Hz, as shown in Figure 4. This suggests that it is possible to distinguish whether EEG data is from a healthy subject or an epileptic patient based on the frequency distribution of frequencies corresponding to each of the multiple dynamic modes of EEG data.

[0071] Fig. 5 shows an example of the discrimination accuracy when EEG data of a healthy subject and EEG data of an epilepsy patient are discriminated based on mode information. In Fig. 5, "Diag" refers to the diagonal elements of the feature matrix (R1 or R2) corresponding to the mode information described above, "OffDiag" refers to the off-diagonal elements of the feature matrix (R1 or R2), and "Fbin" refers to the frequency information included in the mode information.

[0072] As shown in FIG. 5, it was found that the EEG data of healthy subjects and that of epilepsy patients can be distinguished by using at least one of "Diag," "OffDiag," and "Fbin."

[0073] 6 is a diagram showing an example of the results of distinguishing between brain activity of a healthy subject and brain activity of an Alzheimer's patient. Here, the brain activity of the healthy subject is an example of a first event, and the brain activity of the Alzheimer's patient is an example of a second event. The EEG data of the healthy subject is an example of first time-series data, and the EEG data of the Alzheimer's patient is an example of second time-series data.

[0074] In the graph shown in Figure 6, graph 54 shows the frequency distribution of frequencies corresponding to each of multiple dynamic modes of EEG data measuring the brain activity of a healthy subject. On the other hand, graph 55 shows the frequency distribution of frequencies corresponding to each of multiple dynamic modes of EEG data measuring the brain activity of an Alzheimer's patient. This figure also shows the average frequency and 95% confidence interval. As shown in this graph, a significant difference was observed, for example, around 25 Hz. This suggests that it is possible to distinguish whether the EEG data is from a healthy subject or an Alzheimer's patient based on the frequency distribution of frequencies corresponding to each of multiple dynamic modes of the EEG data.

[0075] The table in Figure 6 shows an example of the discrimination accuracy when discriminating between the brain activity of a healthy subject and that of an Alzheimer's patient based on mode information. As shown in Table 6, it was found that by using at least one of "Diag," "OffDiag," and "Fbin," it is possible to discriminate between the EEG data of a healthy subject and that of an epilepsy patient.

[0076] REFERENCE SIGNS LIST 1 Identification device 2 Measurement device 11 Acquisition unit 12 Decomposition unit 13 Identification unit S101 Acquisition step S102 Decomposition step S104 Identification step

Claims

1. An identification method executed by one or more computers, comprising: an acquisition step of acquiring first time series data measuring a first event and second time series data measuring a second event; a decomposition step of obtaining first mode information by decomposing the first time series data into a plurality of dynamic modes and second mode information by decomposing the second time series data into the plurality of dynamic modes; and an identification step of identifying the first event and the second event based on a comparison result between the first mode information and the second mode information.

2. The method of claim 1, wherein the first mode information includes first frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the first time series data, and the second mode information includes second frequency information regarding frequencies corresponding to each of the plurality of dynamic modes of the second time series data, and in the identification step, the first event and the second event are identified based on a comparison result between the first frequency information and the second frequency information.

3. The identification method described in claim 2, wherein the first frequency information is a statistic indicating a frequency distribution of frequencies corresponding to each of the multiple dynamic modes of the first time series data, and the second frequency information is a statistic indicating a frequency distribution of frequencies corresponding to each of the multiple dynamic modes of the second time series data.

4. The identification method described in claim 2, wherein the first frequency information is a number of numerical values ​​indicating frequencies corresponding to each of the multiple dynamic modes of the first time series data belonging to a predetermined frequency band, and the second frequency information is a number of numerical values ​​indicating frequencies corresponding to each of the multiple dynamic modes of the second time series data belonging to the predetermined frequency band.

5. The identification method described in claim 2, wherein the first frequency information is a plurality of numerical values ​​indicating frequencies corresponding to each of the plurality of dynamic modes of the first time series data, and the second frequency information is a plurality of numerical values ​​indicating frequencies corresponding to each of the plurality of dynamic modes of the second time series data.

6. The method of claim 1, wherein the first mode information includes at least one of diagonal components of a feature matrix R1 obtained for the first time series data based on a matrix P1 in which components corresponding to each of the multiple dynamic modes are arranged in a matrix and a matrix Q1 that is a transposed matrix of the matrix P1, and off-diagonal components of the feature matrix R1; and the second mode information includes at least one of diagonal components of a feature matrix R2 obtained for the second time series data based on a matrix P2 in which components corresponding to each of the multiple dynamic modes are arranged in a matrix and a matrix Q2 that is a transposed matrix of the matrix P2, and off-diagonal components of the feature matrix R2.

7. The method of claim 1, wherein the first event and the second event are brain activities of a living organism of interest and a control organism, respectively; the first time series data is electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the living organism of interest; and the second time series data is electroencephalogram data of multiple brain regions measured by electrodes placed at multiple positions on the body of the control organism.

8. The method of claim 1, wherein the first event and the second event are displacement patterns of multiple parts of a moving target object and a moving control object, respectively, the first time series data includes data measuring the displacement over time of each of the multiple parts of the target object, and the second time series data includes data measuring the displacement over time of each of the multiple parts of the control object.

9. An identification device comprising: an acquisition unit that acquires first time series data measuring a first event and second time series data measuring a second event; a decomposition unit that obtains first mode information obtained by decomposing the first time series data into a plurality of dynamic modes and second mode information obtained by decomposing the second time series data into the plurality of dynamic modes; and an identification unit that identifies the first event and the second event based on a comparison result between the first mode information and the second mode information.

10. A control program for causing a computer to function as the identification device according to claim 9, the control program causing a computer to function as the acquisition section, the resolution section, and the identification section.

11. A computer-readable recording medium having the control program according to claim 10 recorded thereon.

Images