Learning model generation device, estimation system, learning model generation method, and program

Abstract

To provide a learning model for shortening a time up to an output of an estimation result in the case of estimating characteristics of an object on the basis of a signal showing a detection result of a sensor.SOLUTION: In a learning model generation device 100, a control part includes: a generation part for generating a learning model for estimating characteristics of an object on the basis of a characteristic amount of a waveform of a signal output from a sensor during a detection period from switching from a first state unexposed to the object to a second state exposed to the object to switching from the second state to the first state; and a preprocessing part for performing preprocessing to the signal before generating the learning model. The preprocessing part has: a setting part for setting an object period of a signal to be used as teacher data of the learning model within the detection period; and a clip part for clipping a signal during the set object period from the signal during the detection period. The generation part generates a learning model by performing machine learning with the characteristic amount of the waveform of the signal during the object period as an explanatory variable and with the characteristics of the object as an objective variable.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a learning model generation device, an estimation system, a learning model generation method, and a program. [Background technology]

[0002] Patent Document 1 describes a method of matching a search object noun, a search object adjective, and a search object numerical value with combination data, and identifying an object having a search object noun that indicates the scent quality of the search object adjective and the search object numerical value based on the combination data. Patent Document 2 describes a method of machine learning the relationship between the value of an estimation target and the corresponding output from outputs from a chemical sensor for multiple samples whose values ​​of a specific estimation target are known. [Prior art document] [Patent documents] [Patent Document 1] Patent No. 7074365 [Patent Document 2] Patent No. 6663151 Summary of the Invention [Problem to be solved by the invention]

[0003] When estimating the characteristics of an object based on a signal indicating the detection result of a sensor, it is desirable to provide a learning model that can shorten the time until the estimation result is output while ensuring that the reliability of the estimation result meets the desired reliability. [Means for solving the problem]

[0004] A learning model generation device according to one aspect of the present invention may include a generation unit that generates a learning model that estimates characteristics of an object based on waveform features of a signal output from a sensor having a sensor unit that physically changes in response to at least one substance emitted from the object and outputs a signal corresponding to the physical change during a detection period from when the sensor switches from a first state in which the sensor is not exposed to the object to a second state in which the sensor is exposed to the object and until when the sensor switches again from the second state to the first state. The learning model generation device may include a preprocessing unit that performs preprocessing on the signal before generating the learning model. The preprocessing unit may include a setting unit that sets a target period of a signal to be used as training data for the learning model within the detection period, and the preprocessing unit may include a clipping unit that clips the signal during the set target period from the signal during the detection period. The generation unit may generate the learning model by machine learning using waveform features of the signal during the target period as explanatory variables and the characteristics of the object as target variables.

[0005] The learning model generation device may further include a period receiving unit that receives input of a target period of a signal to be used as training data for the learning model, and the setting unit may set the target period of the signal to be used as training data for the learning model based on the target period received by the period receiving unit.

[0006] In any of the learning model generation devices, the preprocessing unit may include a reliability receiving unit that receives a reliability required for the learning model. The preprocessing unit may include a derivation unit that derives a reliability of the learning model generated by the generation unit. The preprocessing unit may execute a generation process that causes the setting unit to set a candidate target period, the clipping unit to clip the signal during the candidate target period, and the generation unit to perform machine learning using waveform features of the signal during the candidate target period as explanatory variables and characteristics of the object as objective variables to generate a candidate learning model. The preprocessing unit may further include a determination unit that causes the derivation unit to derive a reliability of the candidate learning model and determines whether to use the candidate learning model as a target learning model based on the reliability received by the reliability receiving unit and the reliability of the candidate learning model.

[0007] In any of the learning model generation devices, the determination unit may determine the candidate learning model as the learning model to be used if the reliability of the candidate learning model is equal to or greater than the reliability accepted by the reliability acceptance unit.

[0008] In any of the learning model generation devices, if the reliability of the candidate learning model is lower than the reliability accepted by the reliability acceptance unit, the determination unit may cause the setting unit to set a new candidate target period longer than the target period of the previous candidate, and execute the generation process for the new candidate target period to generate a new candidate learning model, and further cause the derivation unit to derive the reliability of the new candidate learning model, and if the reliability of the new candidate learning model is equal to or higher than the reliability accepted by the reliability acceptance unit, determine the new candidate learning model to be the learning model to be used.

[0009] In any of the learning model generation devices, the determination unit may cause the setting unit to set target periods for multiple candidates, and execute the generation process for the target periods of the multiple candidates, thereby causing the generation unit to generate learning models for multiple candidates, and further cause the derivation unit to derive reliability of the multiple candidate learning models, and determine that among the multiple candidate learning models, the candidate learning model that satisfies the reliability accepted by the reliability acceptance unit and has the shortest target period is the learning model to be used.

[0010] In any of the learning model generation devices, the preprocessing unit may include a derivation unit that derives the reliability of the learning model generated by the generation unit. The preprocessing unit may cause the setting unit to set a candidate target period, the clipping unit to clip the signal during the candidate target period, and the generation unit to perform machine learning on a plurality of candidate target periods to generate candidate learning models using waveform features of the signal during the candidate target period as explanatory variables and characteristics of the object as objective variables, thereby causing the generation unit to generate a plurality of candidate learning models, and may include a presentation unit that presents a combination of the target period and reliability for each of the plurality of candidate learning models. The preprocessing unit may include a selection receiving unit that receives a selection of a combination of the target period and reliability to be used as a learning model to be used from the combinations of the target period and reliability presented by the presentation unit.

[0011] In any of the learning model generation devices, the preprocessing unit may include a characteristic receiving unit that receives candidate characteristics of an estimation target. The generation unit may generate the learning model that estimates the characteristic of the object from among the candidate characteristics received by the characteristic receiving unit based on the feature amount of the waveform of the signal output from the sensor, using the feature amount of the waveform of the signal output from the sensor as an explanatory variable and each of the candidate characteristics as a target variable.

[0012] In any of the learning model generation devices, the property of the object may be at least one of the type of material and the state of the material.

[0013] In any of the learning model generation devices, the sensitive part may include a sensitive film that is deformed by the at least one substance being adsorbed and diffusing therein.

[0014] In any of the learning model generation devices, the sensitive film may include an organic-inorganic hybrid material.

[0015] An estimation system according to one aspect of the present invention may include the learning model generation device, an acquisition unit that acquires a signal output from the sensor, and an estimation unit that estimates characteristics of the object based on waveform features of the signal and the learning model to be used.

[0016] A learning model generation method according to one aspect of the present invention may include a step of generating a learning model that estimates characteristics of an object based on waveform features of a signal output from a sensor having a sensor unit that physically changes in response to at least one substance generated by the object and outputs a signal corresponding to the physical change during a detection period from when the sensor switches from a first state in which the sensor is exposed to a reference object to a second state in which the sensor is exposed to the object, until when the sensor switches from the second state to the first state again. The learning model generation method may include a step of preprocessing the signal by a preprocessing unit before generating the learning model. The preprocessing step may include a step of setting a target period of a signal to be used as training data for the learning model within the detection period. The preprocessing step may include a step of clipping the signal during the set target period from the signal during the detection period. The generating step may include a step of generating the learning model by machine learning using waveform features of the signal during the target period as explanatory variables and the characteristics of the object as target variables.

[0017] A program according to one aspect of the present invention, when executed by a computer, may cause the computer to function as a generation unit that generates a learning model for estimating characteristics of an object based on waveform features of a signal output from a sensor having a sensor element that physically changes in response to at least one substance emitted from the object and outputs a signal corresponding to the physical change during a detection period from when the sensor switches from a first state in which the sensor is exposed to a reference object to a second state in which the sensor is exposed to the object, until when the sensor switches from the second state to the first state again. When executed by a computer, the program may cause the computer to function as a preprocessing unit that preprocesses the signal before generating the learning model. The preprocessing unit may include a setting unit that sets a target period within the detection period for a signal to be used as training data for the learning model. The preprocessing unit may include a clipping unit that clips signals during the set target period from signals during the detection period. The generation unit may generate the learning model by machine learning using waveform features of the signal during the target period as explanatory variables and characteristics of the object as target variables.

[0018] The above summary of the invention does not list all of the features of the present invention, and subcombinations of these features may also be inventions. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a diagram illustrating an example of functional blocks in the overall configuration of an estimation system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram showing an example of the waveform of a signal (output voltage) output by an olfactory sensor. [Figure 3] This figure shows an example of the results of principal component analysis of the waveforms of the signals from each channel when the olfactory sensor was exposed to white miso, soybean miso (Kyoto), and soybean miso (Echigo) as the target substances, and the conditions were alternated at 1-minute intervals for 15 minutes. The waveforms were taken from the first 10 seconds of the target period. [Figure 4]This figure shows an example of the results of principal component analysis of the waveforms of the signals from each channel when the olfactory sensor was exposed to white miso, soybean miso (Kyoto), and soybean miso (Echigo) as the target substances, and the conditions were alternated at 1-minute intervals for 15 minutes. The waveforms were taken from the first 20 seconds of the target period. [Figure 5] FIG. 2 is a diagram illustrating an example of functional blocks of a learning model generation device. [Figure 6] FIG. 10 is a diagram illustrating an example of a setting screen for inputting conditions for a learning model to be generated by a generation unit. [Figure 7] FIG. 2 is a diagram illustrating an example of functional blocks of the estimation device. [Figure 8] FIG. 10 is a diagram showing an example of an estimation processing screen displayed when the estimation unit estimates the characteristics of an object using a learning model generated by the generation unit. [Figure 9] 10 is a flowchart illustrating an example of a procedure for a learning model generation device to generate a learning model. [Figure 10] FIG. 10 is a diagram for explaining a target period of a signal. [Figure 11] 10 is a flowchart illustrating an example of a procedure for a learning model generation device to generate a learning model. [Figure 12] 10 is a flowchart illustrating an example of a procedure for a learning model generation device to generate a learning model. [Figure 13] FIG. 2 illustrates an example of a hardware configuration. DETAILED DESCRIPTION OF THE INVENTION

[0020] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention according to the claims. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0021] FIG. 1 is a diagram showing an example of functional blocks of the overall configuration of an estimation system 10 according to this embodiment. The estimation system 10 generates a learning model for estimating characteristics of an object based on the detection result of at least one substance generated from the object, and estimates the characteristics of the object using the generated learning model. The object is, for example, food. The estimation system 10 may estimate the type of object as the characteristic of the object. The estimation system 10 may estimate the state of the object as the characteristic of the object. The state of the object may be freshness, decay, fermentation, or aging state, etc.

[0022] The estimation system 10 includes a container 20, a switching mechanism 30, a sensor module 50, a display unit 60, a learning model generation device 100, and an estimation device 200.

[0023] The container 20 contains an object such as food or food ingredients. A switching mechanism 30 is connected to the container 20. The switching mechanism 30 has a pipe 31 communicating with the internal space of the container 20 and a pipe 33 connected to a pipe 32 communicating with the external space of the container 20 via a switching valve 34, the pipe 33 communicating with either the pipe 31 or the pipe 32. The switching valve 34 includes a valve that communicates between the pipe 33 and the internal space of the container 20 or the external space of the container 20. Gas present in the internal space of the container 20 passes through the pipe 31. Gas present in the external space of the container 20 passes through the pipe 32. The gas present in the internal space of the container 20 includes at least one substance generated from the object. The gas present in the external space of the container 20 may be a gas, such as air, used to purge the gas in the pipe 33 so that the olfactory sensor 52 (described below) is not exposed to at least one substance generated from the object. Since the detection results of the olfactory sensor 52 may be affected by temperature or humidity, the external space of the container 20 may be a space equipped with air conditioning equipment that can adjust the temperature and humidity to meet predetermined conditions.

[0024] The sensor module 50 has an olfactory sensor 52, a temperature and humidity sensor 54, and a measurement control unit 56. The measurement control unit 56 controls the measurements of the olfactory sensor 52 and the temperature and humidity sensor 54. The measurement control unit 56 may be configured by a microprocessor such as a CPU or MPU, a microcontroller such as an MCU, or the like.

[0025] The olfactory sensor 52 is provided inside the pipe 33 and detects gas passing through the pipe 33. The configuration of the switching mechanism 30 shown above is merely an example. The switching mechanism 30 may be any mechanism that can switch between a non-detection state (first state) in which the olfactory sensor 52 is not exposed to at least one substance emitted from the target object, and a detection state (second state) in which the olfactory sensor 52 is exposed to at least one substance emitted from the target object. The switching mechanism 30 may have an injection unit that purges gas around the olfactory sensor 52 by injecting air into the inside of the pipe 33.

[0026] The olfactory sensor 52 may be a surface stress sensor. The surface stress sensor has a sensitive part that physically changes in response to at least one substance emitted from the object and outputs a signal corresponding to the physical change. The sensitive part includes a sensitive film that deforms as at least one substance adsorbs and diffuses into its interior. The sensitive film may include an organic-inorganic hybrid material.

[0027] The organic-inorganic hybrid is RSiO from the viewpoint of sensitivity and stability in a humid environment. 3 / 2 where "R" represents an organic functional group.

[0028] In this embodiment, the organic functional group preferably contains one or more aromatic rings (aromatic ring structures). When an aromatic ring is contained, moisture resistance tends to be further improved. The aromatic ring is not particularly limited and can be appropriately selected taking into consideration the application of the sensor, etc., but examples include aromatic hydrocarbon groups such as phenyl, naphthyl, p-tolyl, and biphenyl; substituted aromatic hydrocarbon groups such as 4-chlorophenyl, 4-methoxyphenyl, 4-aminophenyl, and pentafluorophenyl; heterocyclic hydrocarbon groups such as 3-furyl, 3-thienyl, 2-pyridyl, 3-pyridyl, and 4-pyridyl; and metallocenes such as ferrocenyl. The organic functional group in this embodiment may contain one or more of the above-mentioned aromatic rings.

[0029] The temperature and humidity sensor 54 may be connected in series with the olfactory sensor 52 via piping. The temperature and humidity sensor 54 may detect the humidity and temperature of the gas used for purging or at least one substance generated from the object in the container 20, and provide the temperature information and humidity information to the estimation device 200 and the learning model generation device 100.

[0030] The amount or type of gas generated varies depending on the type or state of the material that makes up the object. Therefore, by identifying at least one of the type and amount of gas generated, it is possible to estimate the characteristics of the object, such as its type or state.

[0031] The learning model generation device 100 generates a learning model that the estimation device 200 uses to estimate the properties of an object.

[0032] Fig. 2 shows an example of the waveform of the signal (output voltage) output by the olfactory sensor 52. The signal waveform shown in Fig. 2 shows the waveform of the signal output from the olfactory sensor 52 from the time when the olfactory sensor 52 switches from a non-detection state (first state) in which it is not exposed to an object to a detection state (second state) in which it is exposed to an object at time 0 second, until it switches from the detection state to the non-detection state again after a detection period of 60 seconds.

[0033] When the estimation device 200 estimates the characteristics of an object from the waveform features of such a signal, the reliability of the object characteristics estimated by the estimation device 200 tends to increase as the signal features increase, that is, as the target period within the detection period that the estimation device 200 uses to estimate the object characteristics is longer.

[0034] Figure 3 shows the results of principal component analysis of the waveforms of the signals from each channel when the olfactory sensor 52 was exposed to white miso, soybean miso (Kyoto), and soybean miso (Echigo) as the target objects, using a film surface-type stress sensor consisting of an array of eight types (eight channels) of sensitive films as the olfactory sensor 52, and the conditions were alternated at one-minute intervals for 15 minutes. The first 10 seconds of the signal waveforms were analyzed as the target period.

[0035] Figure 4 shows the results of principal component analysis of the waveforms of the signals from each channel when the olfactory sensor 52 was exposed to white miso, soybean miso (Kyoto), and soybean miso (Echigo) as the target objects, using a film surface stress sensor consisting of an array of eight types (eight channels) of sensitive films as the olfactory sensor 52, and the conditions were alternated at one-minute intervals for 15 minutes. The first 20 seconds of the signal waveforms were analyzed as the target period.

[0036] 3 and 4, classification of white miso, soybean miso (Kyoto), and soybean miso (Echigo) is better when the target period is 20 seconds than when it is 10 seconds. In other words, the longer the target period, the higher the reliability of the object properties estimated by estimation device 200 may be.

[0037] However, the longer the target period used by the estimation device 200 to estimate the properties of the object, the longer the time it takes for the estimation device 200 to estimate the properties of the object. In other words, there is a trade-off between the reliability of the estimation result of the estimation device 200 and the length of the estimation time of the estimation device 200. Which one is given priority depends on the usage mode of the user.

[0038] Therefore, in this embodiment, the learning model generation device 100 generates a learning model by appropriately adjusting the reliability of the estimation result of the estimation device 200 and the length of the estimation time of the estimation device 200 according to the usage manner of the user.

[0039] FIG. 5 is a diagram illustrating an example of functional blocks of the learning model generation device 100. The learning model generation device 100 includes a control unit 110 and a storage unit 120. The learning model generation device 100 may be configured as a computer. The computer may be a personal computer, tablet computer, smartphone, workstation, server computer, general-purpose computer, or a computer system in which multiple computers are connected. Such a computer system is also considered a computer in the broad sense. The computer may be a dedicated computer designed for the learning model generation process of the learning model generation device 100, or may be dedicated hardware realized by a dedicated circuit. The computer may be implemented in a virtual computer environment. When a computer is used, the learning model generation device 100 is realized by the computer executing a program.

[0040] The control unit 110 may be configured by a microprocessor such as a CPU or an MPU, a microcontroller such as an MCU, etc. The control unit 110 has a generation unit 112 and a preprocessing unit .

[0041] The generation unit 112 generates a learning model that estimates the characteristics of an object based on waveform features of a signal output from the olfactory sensor 52 during a target period within a detection period, from when the olfactory sensor 52 switches from a non-detection state in which it is not exposed to an object to a detection state in which it is exposed to an object, until it switches back from the detection state to the non-detection state. The generation unit 112 may generate a learning model that estimates the characteristics of an object from waveform features of a signal by performing machine learning according to a supervised learning algorithm. The generation unit 112 may generate a learning model that estimates the characteristics of an object from a combination of waveform features of multiple signals by performing machine learning according to a supervised learning algorithm. The algorithm may be any algorithm, such as random forest or logistic regression. The algorithm may also be any classification algorithm, such as a neural network, support vector machine, multiple regression analysis, decision tree, or Gaussian process.

[0042] The preprocessing unit 130 performs preprocessing on the signal before the generation unit 112 generates a learning model. The preprocessing unit 130 includes a setting unit 131, a clipping unit 132, a derivation unit 133, a determination unit 134, a period receiving unit 135, a reliability receiving unit 136, a characteristic receiving unit 137, a presentation unit 138, and a selection receiving unit 139.

[0043] The setting unit 131 sets a target period of a signal to be used as training data for a learning model within the detection period. The target period may be from the start of the detection period to any point within the detection period. The target period may be any period within the detection period. In other words, the target period may be any period from the start of the detection period to the end of the detection period.

[0044] The clipping unit 132 clips the signal of the target period set by the setting unit 131 from the signal of the detection period. For example, the clipping unit 132 clips the signal of the target period from the signal of the detection period shown in FIG.

[0045] The generation unit 112 generates a learning model by machine learning using the feature quantities of the signal waveform during the target period as explanatory variables and the characteristics of the object as objective variables. When the estimation device 200 predicts the characteristics of the object using the learning model generated in this way, the amount of information is reduced, thereby shortening the time required to derive an estimation result. Note that the generation unit 112 may perform machine learning using other parameters, such as the temperature and humidity inside the container 20 measured by the temperature and humidity sensor 54, as explanatory variables in addition to the feature quantities of the signal waveform during the target period, and the characteristics of the object as objective variables.

[0046] The period accepting unit 135 accepts input of a target period of a signal to be used as training data for a learning model. The period accepting unit 135 may accept input of the target period of the signal from a user via a setting screen displayed on the display unit 60. The setting unit 131 sets the target period of the signal to be used as training data for a learning model based on the target period accepted by the period accepting unit 135.

[0047] The reliability receiving unit 136 receives the reliability required for the learning model. The reliability receiving unit 136 may receive an input of the reliability of the learning model from the user via a setting screen displayed on the display unit 60.

[0048] The derivation unit 133 derives the reliability of the learning model generated by the generation unit 112. The derivation unit 133 estimates characteristics of an object using the generated learning model for signals during a target period used as training data, and determines whether the estimated characteristics of the object match characteristics of a true object associated with the target period of the signals used as training data. For example, the derivation unit 133 may determine whether a plurality of signals during a target period used as training data match characteristics of a true object, and derive the reliability based on a value obtained by dividing the number of signals whose characteristics match the characteristics of the true object by the total number of signals used in the determination. The derivation unit 133 may derive the value as the reliability. The derivation unit 133 may derive the reliability according to a predetermined function that uses the value as a variable.

[0049] The determination unit 134 executes a generation process in which the setting unit 131 sets a candidate target period, the clipping unit 132 clips the signal in the candidate target period, and the generation unit 112 performs machine learning using the waveform feature of the signal in the candidate target period as an explanatory variable and the characteristics of the object as a target variable, thereby generating a candidate learning model. Furthermore, the determination unit 134 causes the derivation unit 133 to derive the reliability of the candidate learning model, and determines whether to use the candidate learning model as the learning model to be used based on the reliability accepted by the reliability accepting unit 136 and the reliability of the candidate learning model.

[0050] The determination unit 134 may determine the candidate learning model as the learning model to be used if the reliability of the candidate learning model is equal to or higher than the reliability accepted by the reliability accepting unit 136. If the reliability of the candidate learning model is lower than the reliability accepted by the reliability accepting unit 136, the determination unit 134 causes the setting unit 131 to set a new candidate target period that is longer than the target period of the previous candidate by a predetermined period, and causes the generation unit 112 to generate a new candidate learning model by executing a generation process for the new candidate target period. Furthermore, the determination unit 134 causes the derivation unit 133 to derive the reliability of the new candidate learning model, and if the reliability of the new candidate learning model is equal to or higher than the reliability accepted by the reliability accepting unit 136, determines the new candidate learning model as the learning model to be used. If the reliability of the learning model of the new candidate is lower than the reliability accepted by the reliability accepting unit 136, the deciding unit 134 causes the setting unit 131 to set a target period for the new candidate that is longer than the target period of the previous candidate by a predetermined period, and repeats the above process.

[0051] The determination unit 134 may cause the setting unit 131 to set multiple candidate target periods and execute a generation process for the multiple candidate target periods, thereby causing the generation unit 112 to generate multiple candidate learning models. Furthermore, the determination unit 134 may cause the derivation unit 133 to derive the reliability of the multiple candidate learning models, and determine, from among the multiple candidate learning models, the candidate learning model that satisfies the reliability accepted by the reliability accepting unit 136 and has the shortest target period as the learning model to be used.

[0052] Alternatively, the presentation unit 138 may cause the setting unit 131 to set a candidate target period, the clipping unit 132 to clip the signal for the candidate target period, and the generation unit 112 to perform machine learning using waveform features of the signal for the candidate target period as explanatory variables and characteristics of the object as objective variables to generate candidate learning models, thereby executing a generation process for multiple candidate target periods, thereby causing the generation unit 112 to generate multiple candidate learning models. Furthermore, the presentation unit 138 may present a combination of the target period and reliability for each of the multiple candidate learning models. The presentation unit 138 may present a list of the combinations of the target period and reliability for each of the multiple candidate learning models on the display unit 60.

[0053] The selection receiving unit 139 receives a selection of a combination of a target period and reliability to be used as a learning model to be used from among the combinations of the target period and reliability presented by the presentation unit 138. The selection receiving unit 139 may receive a selection of a combination of a target period and reliability to be used as a learning model to be used from among the combinations of the target period and reliability presented by the presentation unit 138 from the user via the display unit 60.

[0054] The estimation device 200 may estimate the characteristics of an object from among multiple predetermined characteristics of the object based on waveform features of the signal output from the olfactory sensor 52. Therefore, the characteristic receiving unit 137 may receive candidate characteristics of the object to be estimated. The memory unit 120 stores, for each characteristic of the object, relationship information indicating the relationship between the waveform features of the signal that can be used as training data and the characteristics of the object. The characteristic receiving unit 137 may receive at least one characteristic of the object from the user via the display unit 60 from among the multiple characteristics of the object stored in the memory unit 120. The characteristic of the object may be the type of substance. The characteristic of the object may be the state of the substance. For example, the characteristic receiving unit 137 may receive the type of food to be classified as the characteristic of the object. The characteristic receiving unit 137 may receive the state of the food to be classified, such as freshness, spoilage, fermentation, or aging, as the characteristic of the object.

[0055] The generation unit 112 may generate a learning model that estimates the characteristics of the object from among the multiple candidate characteristics accepted by the characteristic accepting unit 137 based on the feature quantities of the waveform of the signal output from the olfactory sensor 52, using the feature quantities of the waveform of the signal output from the olfactory sensor 52 as explanatory variables and each of the multiple candidate characteristics as objective variables. The generation unit 112 may further use the temperature information and humidity information from the temperature and humidity sensor 54 as objective variables to generate a learning model that estimates the characteristics of the object.

[0056] 6 is an example of a setting screen for inputting conditions for a learning model to be generated by the generation unit 112. The setting screen 300 includes a field 302 for accepting the characteristics of the object to be estimated, i.e., the object to be classified. The setting screen 300 includes a field 304 for accepting the selection of an algorithm to be used when the generation unit 112 generates a learning model. The setting screen 300 includes a field 306 for accepting the target period of the signal to be used as an explanatory variable. The setting screen 300 includes a field 308 for accepting the reliability required for the learning model to be generated.

[0057] 7 shows an example of functional blocks of the estimation device 200. The estimation device 200 includes a control unit 210 and a storage unit 220. The control unit 210 may be configured by a microprocessor such as a CPU or an MPU, a microcontroller such as an MCU, or the like.

[0058] The estimation device 200 may be configured with a computer. The computer may be a personal computer, a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, or a computer system in which multiple computers are connected. Such a computer system is also a computer in the broad sense. The computer may be a dedicated computer designed for the estimation processing of the estimation device 200, or may be dedicated hardware realized by a dedicated circuit. The computer may be implemented in a virtual computer environment. When a computer is used, the estimation device 200 is realized by the computer executing a program.

[0059] The control unit 210 has an acquisition unit 212 and an estimation unit 214. The acquisition unit 212 acquires signals output from the olfactory sensor 52 during a detection period from when the olfactory sensor 52 switches from a non-detection state, in which it is not exposed to at least one substance emitted from the target, to a detection state, in which it is exposed to at least one substance, until when it switches from the detection state to the non-detection state again.

[0060] The olfactory sensor 52 may have multiple sensitive parts with different characteristics of physical changes in response to at least one substance generated from the target object. The multiple sensitive parts may have multiple sensitive membranes with different characteristics. The olfactory sensor 52 may have multiple channels that output signals from the multiple sensitive parts, respectively.

[0061] For example, if the object is miso, the type and amount of gas generated will vary depending on the type of miso or its fermentation or maturation state. Therefore, the signals output from the multiple channels of the olfactory sensor 52 will also vary. That is, the waveform feature values ​​of each signal will vary depending on the type of miso or its fermentation or maturation state. By preparing a learning model trained using the waveform feature values ​​of the signal as explanatory variables and the type of miso and its fermentation or maturation state as objective variables, the estimation unit 214 can estimate the type of miso and its fermentation or maturation state from the waveform feature values ​​of the signal.

[0062] Therefore, the estimation unit 214 estimates the characteristics of the object based on a learning model capable of estimating the characteristics of the object from feature quantities of the signal waveform and the signal from the olfactory sensor 52. The estimation unit 214 may estimate the characteristics of the object based on a learning model capable of estimating the characteristics of the object from at least two feature quantities of the signal waveform and the signal from the olfactory sensor 52. The feature quantities of the signal waveform may include at least one of the amplitudes of multiple divided waveforms obtained by dividing the signal waveform at predetermined intervals, the sum of the amplitudes of the multiple divided waveforms, the rate of change of the amplitudes of the multiple divided waveforms, the sum of the rate of change of the amplitudes of the multiple divided waveforms, and the average value of the rate of change of the amplitudes of the multiple divided waveforms. The amplitude of the divided waveform corresponds to the signal strength at a time corresponding to the divided waveform of the signal.

[0063] The characteristic of the object may be at least one of the type of the object and the state of the object. The state of the object is a state in which the type or amount of a substance generated from the object changes due to a change in the state of the object. If the object is a food, the state of the object may be the quality state, decay state, maturation state, fermentation state, or ripening state of the food.

[0064] The acquisition unit 212 may acquire a plurality of signals output from the plurality of sensory units during a target period. The estimation unit 214 may estimate the characteristics of the object based on the plurality of signals and a learning model that estimates the characteristics of the object from a combination of waveform features of at least two of the plurality of signals.

[0065] The combination of features of the waveforms of at least two signals may include at least one of the following: a ratio between the amplitudes of each of the multiple divided waveforms of the at least two signals obtained by dividing the waveforms of the at least two signals at a predetermined interval t; a ratio between the sums of the amplitudes of each of the multiple divided waveforms of the at least two signals; a ratio between the rates of change of the amplitudes of each of the multiple divided waveforms of the at least two signals; a ratio between the sums of the rates of change of the amplitudes of each of the multiple divided waveforms of the at least two signals; and a ratio between the average values ​​of the amplitudes of each of the multiple divided waveforms of the at least two signals.

[0066] 8 shows an example of an estimation processing screen 310 that is displayed when the estimation unit 214 estimates the characteristics of an object using a learning model generated by the generation unit 112. The estimation processing screen 310 includes a field 312 that indicates the status of the estimation processing by the estimation unit 214. The status of the estimation processing may be "estimating," "estimation completed," etc. The estimation processing screen 310 includes a field 314 that indicates the classification result as the estimation result by the estimation unit 214. The estimation processing screen 310 includes a field 316 that indicates the waveform of the signal from each channel of the olfactory sensor 52 that is used for estimation by the estimation unit 214.

[0067] FIG. 9 is a flowchart showing an example of a procedure by which the learning model generation device 100 generates a learning model.

[0068] The characteristic receiving unit 137 receives at least one characteristic of an object from the user via the display unit 60 from among the characteristics of a plurality of objects stored in the storage unit 120 (S100). The reliability receiving unit 136 receives an input of a reference reliability of the learning model from the user via a setting screen displayed on the display unit 60 (S102).

[0069] The selection receiving unit 139 receives from the user a selection of an algorithm to be used by the generation unit 112 (S104). The selection receiving unit 139 may receive a selection of either random forest or logistic regression. The determination unit 134 causes the setting unit 131 to set a candidate target period (S106). For example, the determination unit 134 causes the setting unit 131 to set the period from time 0 to time t0 of a signal as shown in FIG. 10 as the target period.

[0070] The generation unit 112 generates a candidate learning model that estimates the characteristics of the target object based on the features of the waveform of the signal output from the olfactory sensor 52 during the target period within the detection period from when the olfactory sensor 52 switches from a non-detection state in which it is not exposed to the target object to a detection state in which it is exposed to the target object, until when it switches from the detection state to the non-detection state again, in accordance with each accepted condition (S108).

[0071] The derivation unit 133 derives the reliability of the candidate learning model generated by the generation unit 112 (S110). The determination unit 134 determines whether the reliability of the candidate learning model generated by the generation unit 112 is equal to or higher than a reference reliability (S112). If the reliability of the candidate learning model is lower than the reference reliability, the determination unit 134 causes the setting unit 131 to set a new candidate target period (S106). The determination unit 134 causes the setting unit 131 to set a new candidate target period that is longer than the previous target period. For example, the determination unit 134 causes the setting unit 131 to set the period from time 0 to time t1 of the signal shown in FIG. 10 as the target period.

[0072] On the other hand, if the reliability of the candidate learning model is equal to or higher than the reference reliability, the determination unit 134 determines the candidate learning model as the learning model to be used (S114).

[0073] Through the above processing, the learning model generation device 100 can provide a learning model that can obtain estimation results that satisfy the standard reliability using signals for a shorter target period.

[0074] FIG. 11 is a flowchart showing an example of a procedure by which the learning model generation device 100 generates a learning model.

[0075] The characteristic receiving unit 137 receives at least one characteristic of an object from the user via the display unit 60 from among the characteristics of a plurality of objects stored in the storage unit 120 (S200). The reliability receiving unit 136 receives an input of a reference reliability of the learning model from the user via a setting screen displayed on the display unit 60 (S202).

[0076] The selection receiving unit 139 receives from the user a selection of the algorithm to be used by the generation unit 112 (S204). The selection receiving unit 139 may receive a selection of either random forest or logistic regression.

[0077] The determining unit 134 causes the setting unit 131 to set a plurality of candidate target periods (S206). For example, the setting unit 131 sets the periods from time 0 to time t0, time t1, time t2, time t3, and time t4 of the signal shown in Fig. 10 as the plurality of candidate target periods.

[0078] The generation unit 112 generates multiple candidate learning models that estimate the characteristics of the target object based on the waveform features of the signal output from the olfactory sensor 52 during multiple target periods within the detection period, in accordance with the received conditions (S208).

[0079] The derivation unit 133 derives the reliability of the multiple candidate learning models generated by the generation unit 112 (S210). The determination unit 134 determines, from the multiple candidate learning models, the candidate learning model whose reliability is equal to or greater than the reference reliability and whose target period is the shortest, as the learning model to be used (S212).

[0080] Through the above processing, the learning model generation device 100 can provide, as a learning model to be used, a learning model that can estimate the characteristics of an object using a signal with the shortest target period that satisfies a desired reliability.

[0081] FIG. 12 is a flowchart showing an example of a procedure by which the learning model generation device 100 generates a learning model.

[0082] The characteristic receiving unit 137 receives at least one characteristic of an object from the user via the display unit 60 from among the characteristics of a plurality of objects stored in the storage unit 120 (S300). The selection receiving unit 139 receives from the user a selection of an algorithm to be used by the generation unit 112 (S302). The selection receiving unit 139 may receive a selection of either random forest or logistic regression.

[0083] The presentation unit 138 causes the setting unit 131 to set a plurality of candidate target periods (S304). For example, the setting unit 131 sets the periods from time 0 to time t0, time t1, time t2, time t3, and time t4 of the signal shown in Fig. 10 as the plurality of candidate target periods.

[0084] The generation unit 112 generates multiple candidate learning models that estimate the characteristics of the target object based on the waveform features of the signal output from the olfactory sensor 52 during multiple target periods within the detection period, in accordance with the received conditions (S306).

[0085] The derivation unit 133 derives the reliability of the multiple candidate learning models generated by the generation unit 112 (S308).

[0086] The presentation unit 138 displays a list of combinations of target periods and reliability levels for each of a plurality of candidate learning models on the display unit 60 (S310). The selection receiving unit 139 receives from the user via the display unit 60 a selection of a combination of target periods and reliability levels to be used as the learning model to be used from among the combinations of target periods and reliability levels presented by the presentation unit 138 (S312). The selection receiving unit 139 determines the selected candidate learning model as the learning model to be used (S314).

[0087] Through the above processing, the learning model generation device 100 can provide, as a learning model to be used, a learning model that satisfies the reliability desired by the user and can estimate the characteristics of an object using signals from a target period desired by the user.

[0088] 13 illustrates an example of a computer 1200 in which aspects of the present invention may be embodied, in whole or in part. A program installed on the computer 1200 may cause the computer 1200 to perform operations associated with an apparatus according to an embodiment of the present invention or to function as one or more “parts” of the apparatus. Alternatively, the program may cause the computer 1200 to perform the operations or one or more “parts.” The program may cause the computer 1200 to perform a process or steps of a process according to an embodiment of the present invention. Such a program may be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0089] The computer 1200 according to this embodiment includes a CPU 1212 and a RAM 1214, which are interconnected by a host controller 1210. The computer 1200 also includes a communication interface 1222 and an input / output unit, which are connected to the host controller 1210 via an input / output controller 1220. The computer 1200 also includes a ROM 1230. The CPU 1212 operates according to programs stored in the ROM 1230 and RAM 1214, thereby controlling each unit.

[0090] The communication interface 1222 communicates with other electronic devices via a network. A hard disk drive may store programs and data used by the CPU 1212 in the computer 1200. The ROM 1230 stores a boot program executed by the computer 1200 upon activation and / or programs dependent on the computer's hardware. The programs may be provided via a computer-readable recording medium such as a CD-ROM, a USB memory, or an IC card, or via a network. The programs may be installed in the RAM 1214 or the ROM 1230, which are also examples of computer-readable recording media, and executed by the CPU 1212. The information processing described in these programs is read by the computer 1200 and establishes cooperation between the programs and the various types of hardware resources described above. An apparatus or method may be configured by implementing operations or processing of information in accordance with the use of the computer 1200.

[0091] For example, when communication is performed between computer 1200 and an external device, CPU 1212 may execute a communication program loaded into RAM 1214 and instruct communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer area provided in RAM 1214 or a recording medium such as a USB memory, and transmits the read transmission data to a network, or writes reception data received from the network to a reception buffer area or the like provided on the recording medium.

[0092] The CPU 1212 may also cause all or a necessary portion of a file or database stored on an external recording medium such as a USB memory to be read into the RAM 1214, and perform various types of processing on the data on the RAM 1214. The CPU 1212 may then write the processed data back to the external recording medium.

[0093] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 1212 may perform various types of processing on data read from the RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 1214. The CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries, each having an attribute value of a first attribute associated with an attribute value of a second attribute, are stored on the recording medium, the CPU 1212 may search for an entry that matches a condition specified by the attribute value of the first attribute from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0094] The above-described programs or software modules may be stored in a computer-readable storage medium on or near the computer 1200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable storage medium, thereby providing the programs to the computer 1200 via the network.

[0095] A computer-readable medium may include any tangible device capable of storing instructions that are executed by an appropriate device. As a result, the computer-readable medium with instructions stored thereon comprises an article of manufacture, including instructions that can be executed to create means for performing the operations specified in the flowchart or block diagram. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable media may include floppy disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disc read-only memory (CD-ROM), digital versatile disc (DVD), Blu-ray (RTM) disc, memory stick, integrated circuit card, etc.

[0096] The computer-readable instructions may include either source code or object code written in any combination of one or more programming languages. The source code or object code may include conventional procedural programming languages. The conventional procedural programming languages ​​may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and the “C” programming language or similar programming languages. The computer-readable instructions may be provided to a processor or programmable circuitry of a programmable data processing apparatus locally or over a local area network (LAN), a wide area network (WAN) such as the Internet, etc. The processor or programmable circuitry may execute the computer-readable instructions to create means for performing the operations specified in the flowcharts or block diagrams.

[0097] Here, the computer may be a computer such as a PC (personal computer), tablet computer, smartphone, workstation, server computer, or general-purpose computer, or may be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system, and is a computer in the broad sense. In a distributed computing system, the multiple computers collectively execute a program by each executing a part of the program and passing data between the computers as needed during program execution.

[0098] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc. A computer may have one or more processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute the program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at time slice intervals. In this case, which portion of a program each processor executes changes dynamically. Alternatively, which portion of a program each of the multiple processors executes may be statically determined by multiprocessor-aware programming.

[0099] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0100] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order. [Explanation of symbols]

[0101] 10 Estimation System 20 containers 30 Switching mechanism 31, 32, 33 Piping 34 Switching valve 50 Sensor Module 52 Olfactory sensor 54 Temperature and humidity sensor 56 Measurement control section 60 Display section 100 Learning model generation device 110 control section 112 Generation part 120 Storage section 130 Pre-processing section 131 Setting section 132 Clip part 133 Derivation part 134 Decision Section 135 Period Reception Department 136 Trust Reception Unit 137 Characteristics Reception Department 138 Presentation section 139 Selection Reception Department 200 Estimation device 210 Control Unit 212 Acquisition Department 214 Estimation Department 220 Storage section 300 Settings screen 310 Estimation processing screen 1200 Computer 1210 host controller 1212 CPU 1214 RAM 1220 Input / Output Controller 1222 communication interface 1230 ROM

Claims

1. a generation unit that generates a learning model for estimating characteristics of an object based on feature quantities of a waveform of a signal output from a sensor having a sensitive part that physically changes in response to at least one substance generated from the object and outputs a signal corresponding to the physical change, during a detection period from when the sensor switches from a first state in which the sensor is not exposed to the object to a second state in which the sensor is exposed to the object until when the sensor switches from the second state to the first state again; a preprocessing unit that performs preprocessing on the signal before generating the learning model; Equipped with The pre-treatment unit a setting unit that sets a target period of a signal to be used as training data for the learning model within the detection period; a clipping unit that clips the signal in the set target period from the signal in the detection period; and The generation unit generates the learning model by machine learning using features of the waveform of the signal in the target period as explanatory variables and characteristics of the object as objective variables.

2. a period receiving unit that receives an input of a target period of a signal to be used as training data for the learning model; The learning model generation device according to claim 1 , wherein the setting unit sets a target period of a signal to be used as training data for the learning model based on the target period accepted by the period accepting unit.

3. The pre-treatment unit a reliability reception unit that receives a reliability required for the learning model; a derivation unit that derives the reliability of the learning model generated by the generation unit; a determination unit that causes the setting unit to set a candidate target period, causes the clipping unit to clip the signal in the candidate target period, and executes a generation process that generates a candidate learning model by performing machine learning on the generation unit using features of the waveform of the signal in the candidate target period as explanatory variables and characteristics of the object as objective variables; and further causes the derivation unit to derive a reliability of the candidate learning model, and determines whether the candidate learning model should be used as a learning model based on the reliability received by the reliability receiving unit and the reliability of the candidate learning model. The learning model generation device according to claim 1 , further comprising:

4. the determination unit determines the candidate learning model as a learning model to be used when the reliability of the candidate learning model is equal to or greater than the reliability accepted by the reliability acceptance unit; The learning model generation device of claim 3, wherein, when the reliability of the candidate learning model is lower than the reliability accepted by the reliability acceptance unit, the determination unit causes the setting unit to set a new candidate target period longer than the target period of the previous candidate, and executes the generation process for the new candidate target period, thereby causing the generation unit to generate a new candidate learning model, and further causes the derivation unit to derive the reliability of the new candidate learning model, and when the reliability of the new candidate learning model is equal to or higher than the reliability accepted by the reliability acceptance unit, determines the new candidate learning model to be the learning model to be used.

5. The learning model generation device of claim 3, wherein the determination unit causes the setting unit to set target periods for multiple candidates, and executes the generation process for the target periods for the multiple candidates, thereby causing the generation unit to generate learning models for multiple candidates, and further causes the derivation unit to derive reliability of the learning models for the multiple candidates, and determines, from among the multiple candidate learning models, the candidate learning model that satisfies the reliability accepted by the reliability acceptance unit and has the shortest target period as the learning model to be used.

6. The pre-treatment unit a derivation unit that derives the reliability of the learning model generated by the generation unit; a presentation unit that causes the setting unit to set a candidate target period, causes the clipping unit to clip the signal in the candidate target period, and causes the generation unit to perform machine learning to generate a candidate learning model by using features of the waveform of the signal in the candidate target period as explanatory variables and characteristics of the object as objective variables, for a plurality of candidate target periods, thereby causing the generation unit to generate a plurality of candidate learning models, and presents a combination of the target period and reliability of each of the plurality of candidate learning models; a selection receiving unit that receives a selection of a combination of a target period and a reliability level to be used as a learning model from among the combinations of the target period and the reliability level presented by the presenting unit; The learning model generation device according to claim 1 , further comprising:

7. The pre-treatment unit The method further includes a characteristic receiving unit that receives a characteristic of a candidate for estimation, 2. The learning model generation device according to claim 1, wherein the generation unit generates the learning model that estimates the characteristics of the object from among a plurality of candidate characteristics accepted by the characteristic accepting unit based on the features of the waveform of the signal output from the sensor, using the features of the waveform of the signal output from the sensor as explanatory variables and each of a plurality of candidate characteristics as a target variable.

8. The learning model generation device according to claim 1 , wherein the property of the object is at least one of a type of material and a state of material.

9. The learning model generation device according to claim 1 , wherein the sensitive part includes a sensitive film that is deformed by the at least one substance being adsorbed and diffusing therein.

10. The learning model generating device according to claim 9 , wherein the sensitive film includes an organic-inorganic hybrid material.

11. A learning model generation device according to any one of claims 1 to 10; an acquisition unit that acquires a signal output from the sensor; an estimation unit that estimates characteristics of the object based on a feature amount of the waveform of the signal and the learning model of the object to be used; An estimation system comprising:

12. a generation unit generating a learning model for estimating characteristics of the object based on feature quantities of a waveform of a signal output from the sensor, the sensor having a sensitive part that physically changes in response to at least one substance generated from the object and outputs a signal corresponding to the physical change, during a detection period from when the sensor switches from a first state in which the sensor is exposed to a reference object to a second state in which the sensor is exposed to the object, until when the sensor switches from the second state to the first state again; a preprocessing unit performing preprocessing on the signal before generating the learning model; Equipped with The pre-treatment step includes: setting a target period of a signal to be used as training data for the learning model within the detection period; clipping the signal during the set target period from the signal during the detection period; and A learning model generation method, wherein the generating step includes a step of generating the learning model by machine learning using features of the waveform of the signal in the target period as explanatory variables and characteristics of the object as objective variables.

13. When executed by a computer, the computer is a generation unit that generates a learning model for estimating characteristics of an object based on feature quantities of a waveform of a signal output from a sensor having a sensitive part that physically changes in response to at least one substance generated from the object and outputs a signal corresponding to the physical change, during a detection period from when the sensor switches from a first state in which the sensor is exposed to a reference object to a second state in which the sensor is exposed to the object, until when the sensor switches from the second state to the first state again; a preprocessing unit that performs preprocessing on the signal before generating the learning model; It functions as The pre-treatment unit a setting unit that sets a target period of a signal to be used as training data for the learning model within the detection period; a clipping unit that clips the signal in the set target period from the signal in the detection period; and The generation unit generates the learning model by performing machine learning using features of the waveform of the signal in the target period as explanatory variables and characteristics of the object as objective variables.

Images