Patent Application
20230090065
The present invention relates to selecting, from among an odor set consisting of a plurality of kinds of odors, a subset of the odor set, which is a subset consisting of a relatively small number of odors and represents any odor in the odor set in an at least approximate manner by using the combination thereof (hereinafter, each individual odor in this subset is referred to as a primary odor). The present invention also relates to representing, presenting, or synthesizing many kinds of odors in an at least approximate manner by using the combination of the primary odors thus selected. The present invention further relates to representing or presenting many kinds of odors in a form perceivable by a perception other than olfaction, for example, in color. The present invention further relates to an apparatus therefor.
As is well known, any color can be separated into three primary colors and represented by using weights for the respective primary colors. Furthermore, any color can be synthesized so as to be restored on human vision by mixing the light of the respective primary colors with the intensity of the weight corresponding to the color to be represented. In principle, such synthesis requires to prepare only three kinds of colors instead of preparing the light emission source or the colorant of every color. Drawing from the analogy of separating or synthesizing a color into or from primary colors as described above, it is conceivable that any odor is separated into a relatively small number of odors, namely, “primary odors” corresponding to primary colors, and any odor is synthesized by mixing such primary odors with weight. If such separation and synthesis of odor is achieved, a huge number of applications will be expected.
Note that, as for the primary odor, separating an odor into the primary odors or synthesizing an odor by mixing some primary odors does not mean separation or synthesis in the sense of the component composition of odor. It should be noted that the primary odor means an original odor from which an odor identical or very similar to any target odor in terms of olfaction (human or other animal olfaction, or more generally response of various odor sensors) can be synthesized by mixing themselves. For example, even if an odor X of gas containing acetic acid can be synthesized by mixing a primary odor A and a primary odor B, it does not necessarily mean that acetic acid is contained in any of the primary odors A and B. It merely means that gas obtained by mixing these primary odors at a certain ratio is perceived as an odor identical or similar to the odor X.
Furthermore, when an odor is converted into a form in which the odor can be recognized by a kind of perception other than olfaction by separating the odor into the primary odors, the odor can be represented more in an easy-to-understand manner. Consider an application in a case, for example, where an odor can be represented by color. If odor can be colored, the effect of “odor” can be used in comics and movies. In addition, if perfumes, cosmetics, whiskey, wine, and the like can be represented by colors, it is possible to visually understand the odor of a product and contribute to sales by adopting the colors in a package of the product. Since color has an effect of influencing human mind, applications are limitless if odor can be represented by color. Furthermore, since a method for representing odor itself has not been established so far, odor can be represented more objectively than before by representing odor as color. It is expected that this makes it easier to store, learn, and understand odor.
However, it is generally difficult to represent odor by separation of odor into the primary odors as described above, synthesize and generate odor by mixture, and convert odor into a form that can be recognized by other perception. This is because the primary odors in olfaction corresponding to the three primary colors of light and the four primary tastes (some say that the number of primary tastes is five, six, or more) is not scientifically identified. Studies for trying to define the primary odor have been conducted for a long time, and Henning's odor prism and the like are famous. Henning's odor prism is proposed by Henning, and in the odor prism, six kinds of flowery, fruity, putrid, spicy, resinous, and burnt are regarded as primary odors. However, when the sample as a primary odor is collected, not all odors can be represented by a mixed odor of these primary odors. In case of the light or taste, any other light or taste can be reproduced by preparing a reference sample, but odor cannot. For example, when it is desired to represent an odor by a single color, three primary odors corresponding to three primary colors of light are required to be found, but from the above reason, it is substantially impossible to determine the three primary odors.
By the way, development of an artificial olfactory system using a sensor array is remarkable in recent years. By using such an artificial olfactory system, it is becoming possible to numerically represent odor. Further, studies for visualizing odor using a sensor array have also been conducted by using various methods, and odor is generally represented as a set of a plurality of colors. Among these studies, use of an optoelectronic nose is the mainstream. In this method, materials, such as a plurality of dyes and pigments, whose colors change when the odor component comes into contact with the materials are prepared, and changes in color when an odor to be targeted is blown onto the materials are detected. However, in these strategies, the primary odor has not been investigated, and odor cannot be represented by a single color.
For example, Non Patent Literature 1 discloses “absolute value expression analysis” in which a comparison result with several kinds of reference gasses defined in advance is quantified and expressed on a radar chart or the like, and an example of such quantification and the radar chart expression is disclosed in Non Patent Literature 2. However, in the absolute value expression described in the above literature, it is necessary for the provider or the user of the measurement system to preset the reference odor (reference gas), but whether or not the odor to be measured can be favorably approximated by using the combination of these reference gasses has not been examined. Therefore, the absolute value expression analysis disclosed in the above literature does not give any suggestion about the above-described odor representation by the separation of odor into the primary odors, synthesis and generation by mixture, and conversion into a form that can be recognized by other perception.
The problem to be solved by the present invention is to systematically select primary odors from a set of odors. Another problem to be solved by the present invention is to separate, synthesize/generate any odor by using the selected primary odors. Still another problem to be solved by the present invention is to represent or present odor by a single color by using the selected primary odors. Yet another problem to be solved by the present invention is to produce a device capable of converting odor into color in real time.
According to one aspect of the present invention, there is provided a method for selecting primary odors comprising: detecting each of odors in a set of the odors; and selecting a subset consisting of primary odors from the set of the odors, based on detection results of each of the odors.
The detecting each of the odors may include obtaining a feature vector corresponding to each of the odors, each of the feature vectors comprising a plurality of features obtained from output signals of a sensor for each of the odors, and the selecting the subset consisting of the primary odors includes selecting the primary odors from a set of the feature vectors respectively corresponding to the odors in the set of the odors, based on a subset of the feature vectors selected by endpoint detection in feature space including the feature vectors.
Each row in a feature matrix may be subjected to scale conversion, the feature matrix being constituted by taking the feature vectors obtained corresponding to the respective odors as column vectors thereof.
The subset of the feature vectors may be obtained by counting the number of times the respective feature vectors are end points, the end point being a point located at an extreme end in any arbitrary direction in the feature space, and extracting the feature vectors from the set of the feature vectors in descending order of the counted number of times.
The sensor may be a surface stress sensor.
The sensor may be a sensor array having a plurality of sensor elements, and the output signals may include individual output signals from the respective plurality of sensor elements.
According to another aspect of the present invention, there is provided an apparatus for selecting primary odors comprising: a sensor; a gas supply means configured to alternately supply sample gases and reference gas to the sensor, each of the sample gases having an odor corresponding to each of odors in a set of the odors; and an information processing device configured to input output signals from the odor sensor, wherein the apparatus for selecting primary odors may be configured to select some odors of the sample gases as primary odors by using the above-described method for selecting primary odors, each of the some odors selected corresponding to each element of the subset of the feature vectors obtained by the information processing device.
According to still another aspect of the present invention, there is provided a method for synthesizing an odor comprising synthesizing, based on a plurality of primary odors selected by using the above-described method for selecting primary odors, a given odor as an odor obtained by mixing the primary odors selected.
According to yet another aspect of the present invention, there is provided an apparatus for synthesizing an odor comprising a means configured to mix the sample gases corresponding to each of the primary odors selected by using the apparatus for selecting primary odors.
According to yet another aspect of the present invention, there is provided a method for representing an odor comprising representing, based on a plurality of primary odors selected by using the above-described method for selecting primary odors, a given odor as a combination of the primary odors selected.
The combination of the primary odors may be represented by a linear combination of feature vectors of the primary odors.
According to yet another aspect of the present invention, there is provided an apparatus for representing an odor comprising: a sensor; a gas supply means configured to alternately supply sample gases and reference gas to the sensor, each of the sample gases having an odor corresponding to each of odors in a set of the odors; an information processing device configured to input output signals from the odor sensor; and a means configured to store a result of odor representation obtained by the information processing device by using the above-described method for representing odor, or configured to provide the result of odor representation to an outside.
According to yet another aspect of the present invention, there is provided a method for presenting an odor comprising presenting the combination of the primary odors in a form recognizable by perception other than olfaction.
The presenting the combination of the primary odors in the form recognizable by the perception other than the olfaction may be performed by associating different colors respectively with the primary odors and presenting the odor by a result of mixing the different colors.
The different colors with which the primary odors are respectively associated may be three primary colors.
According to yet another aspect of the present invention, there is provided an apparatus for presenting an odor comprising: a sensor; a gas supply means configured to alternately supply sample gases and reference gas to the sensor, each of the sample gases having an odor corresponding to each of odors in a set of the odors; an information processing device configured to input output signals from the odor sensor; and an output device configured to output a result of odor presentation obtained by the information processing device by using the above-described method for presenting odor.
According to the present invention, sample gas that may contain a plurality of chemical substances, that is, an odor (gas generated from the sample when the sample to be given is a liquid or a solid) is analyzed, and samples to be served as primary odors, that is, odors are selected from among the sample gasses. By using the primary odors thus selected, any odor can be represented by the combination of the primary odors. This representation may be one that can be recognized by human perception, such as color (e.g., presentation in a form that can be recognized by perception other than olfaction). Moreover, any odor can also be synthesized at least in an approximate manner.
In an embodiment of the present invention, from a set of a plurality of sample gasses (hereinafter, sample gas is also often referred to as “odor”), a subset of these odors is selected by using a nanomechanical sensor and machine learning (When the sample to be given is a liquid or a solid, gas generated therefrom is referred to as sample gas.). The individual odors in the subset of the odors thus selected are each also referred to as a primary odor in the present application. It should be noted that the primary odor is not given a priori (i.e., not given in advance before the application of the present invention) and is dynamically determined in the process of carrying out the present invention, as apparent from the above description. It is a matter of course that once selected primary odors can be reused. For example, in the first measurement, primary odors are selected from a plurality of odors as a subset thereof, and in the subsequent measurement, the primary odors selected in the first time may be used again instead of selecting the primary odors each time. By determining the primary odor in this manner, any odor in a set of given odors can be represented by the combination of odors each serving as a primary odor. The representation of any odor by the primary odors may be provided as mere data (for example, data in which a primary odor 1, a primary odor 2, . . . a primary odor n corresponds to c1%, c2%, . . . and cn %, respectively), or may be converted such that the odor can be recognized by perception other than olfaction in order to represent the odor more intuitively. As an example for the latter case, by respectively assigning different colors to primary odors, odor can be represented as a color obtained by mixing the colors respectively assigned to these primary odors. Typically, after selecting three primary odors, three primary colors can be assigned to these primary odors. Furthermore, not only the odor is represented as a combination of the primary odors, but also an odor identical or similar to the odor capable of being represented by using the combination of the primary odors can be synthesized by mixing these primary odors.
In other words, the present invention provides a novel and practical technical idea that can be realized even in a current situation where “true primary odor” has not been found. In this technical idea, the whole of almost infinite information of all odors (response of a sensor to all odors) is not targeted, but a computable partial range is cut out from the whole information, and the primary odor is logically defined in this range. Furthermore, even if a true primary odor can be identified, a substance of the primary odor that provides the true primary odor is not necessarily easy to obtain and handle, and is not necessarily safe for a living body. In addition, it is possible that the number of kinds of the primary odors is extremely large. Therefore, it is considered that separation of odor into the primary odors, based on the true primary odors cannot be a realistic solution. The present invention is also sufficiently applicable to such a case.
Describing an embodiment of the present invention more specifically, first, a plurality of sensors that exhibit different responsiveness to various odors are used in order to convert an odor into response signals that can be processed numerically. Examples of such a sensor include a nanomechanical sensor that detects a change in a physical parameter on a predetermined surface thereof. One of the nanomechanical sensors is a surface stress sensor. The surface stress sensor is provided with a film (sensitive film) that adsorbs a specimen such as gas on the surface, and expansion and contraction caused in the sensitive film by adsorption appear as a change in surface stress of the surface. Since various patterns of surface stress changes are measured by using the combination of a substance to be measured and a material of a sensitive film (sensitive film material), qualitative or quantitative analysis of various substances in a sample can be performed by supplying a sample to one or a plurality of surface stress sensors and measuring responses thereof.
Although a variety of sensors for detecting odor have been proposed and produced, a nanomechanical sensor using electrical readings based on piezoresistance, which is referred to below as an MSS (Patent Literature 1 and Non Patent Literature 3), will be used as an example of such a sensor. This sensor has several advantages that have been reported as compared with known cantilever-based nanomechanical sensors and achieves high sensitivity, compactness of its system, and high stability. In addition, the size of the sensor channel is smaller than 1 mm, and a plurality of channels coated with different sensitive film materials can be easily prepared. This makes it possible to acquire a lot of information of the odor at a time, and reliable response signals regarded as a “fingerprint” of the odor sample can be collected. Achievements made by use of the feature value extracted from the response signals (hereinafter, a vector defined by a plurality of feature values is referred to as a feature vector) have been observed in quantitative analysis of odor and odor identification. Of course, other kinds of sensors can also be used.
In order to find a primary odor, first, odor samples are collected and converted into respective response signals by using a plurality of MSS in which various sensitive films are used. Then, a high-dimensional feature vector is extracted from the response signals by applying physicochemical knowledge. Next, by executing endpoint detection using machine learning in high-dimensional feature space, ranking of candidates for primary odors is calculated for all the collected odor samples. The number of odors to be selected as primary odors (the number of primary odors) is determined, and the determined number of primary odors are selected in descending order of the ranking. The other odors are each a mixture of these primary odors. If the target odor can be represented or approximated by a linear combination of feature vectors of these primary odors, coefficients of the linear combination can be calculated by, for example, quadratic programming. In this way, any odor can be represented (approximated) as an odor in which a small number of primary odors are mixed.
General description for the “endpoint detection” mentioned above will be provided. The endpoint detection in the present application means any data analysis method for selecting a sample positioned at an end in the feature space. Various kinds of such machine learning algorithms have been proposed (e.g., Non Patent Literature 4). In the present application, as an example of such a method, a method of finding points located at ends in various directions of the feature space, counting the number of appearances for each point, selecting points in descending order of the number of appearances, and selecting odors associated with these points as primary odors will be described below by applying the method according to Non Patent Literature 5.
In the field of machine learning, the “end” is referred to as an endpoint or an endmember. In accordance with this term, in the present application, a point located at an end of the feature space is defined as an endpoint, and a data analysis method for detecting the endpoint is referred to as endpoint detection. As described above, there are various data analysis methods for detecting an endpoint, and in the present application, a specific method described below is exemplified as one example. However, it should be noted that the endpoint detection is not limited to the specific method, and any data analysis method for detecting an endpoint can be generally used.
In addition to dynamically determining which odor is selected as a primary odor, the number of primary odors may also be dynamically set. For example, in the case where the degree of approximation of an odor other than the primary odors, which is synthesized when a certain number of primary odors are selected, is lower than an expected value, the accuracy of approximation can also be improved by sequentially adding, to the primary odors, higher-order odors that are listed in the above-described ranking and have not been selected, instead of setting the number of primary odors to a preset fixed number. That is, by determining the intensity (weight) of each primary odor and mixing them, an odor other than the primary odors can be synthesized (approximately in many cases). Such synthesis can be implemented by actually mixing the primary odors based on the ratio of the above weight. Alternatively, instead of actually synthesizing an odor, an odor other than the primary odors can be represented on the data as a combination of the selected primary odors. Such representation can be implemented by numerically approximating the feature vector of the odor to be approximated as a linear combination of the primary odors. Examples of such approximation include reconstructing the feature vector of the odor to be approximated as a linear combination of the feature vectors of the primary odors. The degree of approximation between the synthesized or approximated odor and the target to be synthesized or approximated, that is, the target odor, may be determined by a sensory test. Alternatively, a difference between feature vectors of both odors or a difference between response signals themselves of the sensors may be calculated.
The number of primary odors can be set to any number in accordance with the purpose of use of the primary odor. For example, when an odor is displayed in association with a color, the top three samples in the ranking are defined as three primary odors. These primary odors are assigned to three primary colors of red, green, and blue, respectively. However, it is not essential to use three primary colors as colors with which the primary odors are respectively associated, and any three colors can be assigned as the colors for the three primary odors. In this case, the position on the chromaticity coordinate of the target odor is calculated, and the color for the target odor is acquired from the position on the chromaticity coordinate. An apparatus for outputting an odor in a single color in real time can be produced by using, for example, an MSS, a compact computer that implements a machine learning method, an LED light, or the like.
A specific method for selecting primary odors will be described below.
[Feature Value Extraction from Response Signals]
Odors are measured by a sensor array which is an aggregate of MSS (hereinafter, each MSS is also referred to as a channel) coated with different sensitive film materials as sensitive films. In a surface stress sensor such as an MSS, a sample to be measured and a reference fluid (Reference gas in the case of odor measurement. When the odor to be measured is supplied to the sensor, the component in the odor is adsorbed to the sensitive film, but the adsorbed component is desorbed while the reference gas flows, and thus reference gas is also called as purge gas in this sense. In the examples, nitrogen gas (N2) is used.) is alternately supplied at a predetermined period, thereby obtaining response signals based on periodic adsorption and desorption occurring in the sensitive film.
A conceptual diagram showing an example of a configuration of a measurement apparatus for performing such measurement is given in
More specifically describing such information processing, first, the response signals as shown in
The shape of the above peak is not actually constant, and may gradually change from immediately after the start of alternate supply of the sample gas (odor) and the reference gas. Therefore, as for an example of the peak from which the parameter is extracted, it may be preferable to appropriately select a peak that appears after the alternate supply is repeated a sufficient number of times from the start of the alternate supply.
In addition, each sample gas (odor) may be gas from the beginning. Alternatively, the odor generation source may be a liquid or a solid. In the latter case, gas evaporated from the liquid or the solid may be set to a target to be measured. In the embodiment, the odor generation source is described as a liquid, but this description does not lose generality.
In order to determine a set of primary odors from the collected odor samples, the following endpoint detection method is performed on a high-dimensional feature vector of the odor obtained by measurement using the MSS. Of course, as described above, the endpoint detection method described below is merely an example, and it goes without saying that any other method can be adopted. To simplify the description below, it is assumed that the number of primary odors is three, and the feature values extracted from each odor sample are the parameters 1 to 4 described above with reference to
It is assumed that N is the number of odor samples and
X
i [Mathematical Formula 1]
is a column vector including a 48-dimensional feature of the i-th odor sample. At this time, the feature matrix X is defined as follows.
X(x1,x2, . . . ,xN) [Mathematical Formula 2]
Each row of the feature matrix X is standardized such that the mean and the standard deviation are 0 and 1, respectively.
Expressed by using a mathematical expression, the above-described standardization is processing in which, for values of each row of the above-described feature matrix xj1, xj2, . . . , xji, . . . , xjN, xji is converted into (xji−μ)/σ. In the above expression, the subscripts i and j and the constants μ and σ are defined as follows.
This standardization means that each feature vector is translated in the feature space and distributed around the origin of the feature space by setting the mean of each row of the feature matrix to 0. Further, setting the standard deviation of each row to 1 means that the “effectiveness” of individual MSS and parameters, that is, the degree of contribution of individual MSS and parameters to the selection of the primary odor by aligning the MSS to be used and the sizes and variations of the feature values (parameters) obtained from these MSS.
Note that setting the mean of each row to 0 in the above-described standardization does not essentially affect endpoint detection. However, in the machine learning field, standardization in which the mean is set to 0 and the standard deviation is set to 1 (such standardization is also referred to as z-score) is often performed, and thus this standardization is also adopted in this case. Of course, instead of standardization, it is possible to use a normalized value of each row or use the value of each row as it is, and the standardization is not necessarily required. Standardization means to set variance to 1 and mean to 0, and normalization means to convert each row to set the maximum to 1 and the minimum to 0. Additionally, it is not necessarily intended that scale conversion as part of data preprocessing is limited to standardization or normalization, and any scale conversion processing can be applied in an appropriate manner. It should also be noted that such scale conversion is not essential, as a matter of course.
In the endpoint detection method, first, a value K as large as possible (depending on the required accuracy and the like of selection of primary odors, 200 times or more the number of feature values is desirable, for example) is set, and a set of K random unit vectors
{sk}k=1, . . . ,K [Mathematical Formula 3]
is generated. Being “random” means having a randomness that is uniform with respect to directions in feature space (in this case, 48-dimensional vector space). The dimension of each unit vector
S
k [Mathematical Formula 4]
is identical to the dimension of the feature vector, and is 48 dimensions in this case.
Also, a vector
n={n
i=0}i=1,2, . . . ,N [Mathematical Formula 5]
is prepared. This vector is used to store a score related to the endpoints, called the endpoint score.
In this endpoint detection method, the following procedure is repeated K times.
(1) Feature vectors of all standardized odor data
X
i [Mathematical Formula 6]
are projected onto each unit vector
S
k [Mathematical Formula 7]
That is, a vector having, as an element, an inner product of the standardized feature vector of each odor
X
i [Mathematical Formula 8]
(wherein, i=1 to N)
and a given unit vector
S
k [Mathematical Formula 9]
is calculated as follows:
y=X
T
s
k [Mathematical Formula 10]
This means coordinates in directions of a unit vector
S
k [Mathematical Formula 11]
of the feature vector for each of the odor samples 1 to N. (Note that the feature vector starts at the origin of the feature space.)
(2) Indices I− and I+ of the minimum and maximum elements among elements (coordinates) in a vector
y [Mathematical Formula 12]
are acquired as follows.
(3) For the indices I− and I+ of selected element, a component of vector
n [Mathematical formula 14]
is updated as follows. That is, the endpoint score of the element having the selected index is increased by 1.
n
1−
→N
I−+1,
n
1+
→N
I++1 [Mathematical Formula 15]
An example of this iterative procedure is shown in
The method for selecting a plurality of primary odors (as an example, three primary odors in the above description; hereinafter, description will be made based on assumption that the number of primary odors is three) among odor samples has been described above in [Endpoint detection method for selecting primary odors]. In order to express another odor by the combination of the primary odors, the feature vector of the other odor is represented by a linear combination of the feature vectors of the three primary odors. At this time, a linear combination in which a feature vector of an odor to be targeted is realized as much as possible is searched. It is assumed that (w1, w2, w3) is coefficients representing a linear combination of a standardized feature vector of three primary odors
X
e
,X
e
, and Xe
In the following description, one of the three primary colors of color is associated with corresponding one of the three primary odors, whereby any odor is represented as a mixed color of the three primary colors. However, this description does not also lose generality as a matter of course.
Coefficients representing the linear combination are obtained by executing a quadratic programming. That is, the standardized feature vector of the odor to be targeted is assumed to be
X [Mathematical formula 17]
coefficients (w1, w2, w3) that minimize a difference A defined by the following formula are obtained.
Provided, however, that the following two conditions are imposed.
The former condition is required to represent the target odor as a positive mixture of three primary odors. This originates from that the primary odor cannot be mixed at a negative concentration. On the other hand, the latter condition is a condition necessary for associating wi with the concentration at which the primary odor is mixed. When another odor is represented in color, the position of the chromaticity triangle of the target odor is given as follows.
(w1/2+w2,√{square root over (3)}w1/2) [Mathematical Formula 20]
The color associated with the target odor is determined by using the color in the chromaticity triangle corresponding to this position. In this case, the mixed concentration of the primary odor is determined such that the standardized feature vector is approximated, but the target to be approximated is not limited thereto. A normalized feature vector, or a feature vector that is not standardized may be employed. Alternatively, response signals may be employed directly.
Then, the top three samples serving as the candidates for primary odor are regarded as three primary odors, and are used as vertices of a color triangle as in three primary colors of colors, thereby representing all odors with color. As an example of the sample, pure water and seasonings are used in examples. However, it will be obvious from the following description that these samples and targets to be measured are merely examples, and various other kinds of samples and chemical substances can be freely selected. Furthermore, it will be also obvious that the present invention can also be applied to a liquid or solid sample that generates odor.
[Feature Extraction from Response Signals and Used Sensitive Film Materials]
The parameters 1 to 4 that are described above with reference to
Silica/titania composite nanoparticles subjected to various surface functionalization, various polymers, and SiO2-C16TAC (silica/hexadecyltrimethylammonium composite particles) were prepared as 12 kinds of sensitive film materials. These were applied to the surface of the MSS by using several coating techniques. The material applied to each channel and the coating method are stated in [Details of examples—production of MSS chip coated with various materials] to be described later.
As described above, one odor sample is represented by 48 parameters (product of four features and 12 channels). This serves as a high-dimensional feature value for each odor sample, that is, a feature vector.
[Selection of Primary Odors from Odors of Pure Water and Seasonings]
In order to show an example, the primary odors are selected from the odors of pure water and seasonings using the MSS and machine learning. In this case, twelve kinds of seasonings were prepared, namely, pure water, ketchup, mayonnaise, lemon juice, oyster sauce, Worcestershire sauce, Japanese rice wine, Japanese noodle sauce, grilled meat sauce, grain vinegar, soy sauce, and fish sauce were used. Each of these liquid samples was put in a separate vial, and the gas accumulated in the headspace in the vial was measured by MSS. In the present embodiment, the N2 gas identical to the reference gas was used as carrier gas for pushing out the gas (odor) accumulated in the headspace and sending the gas to the MSS. The dependency of the response signals with respect to the sensitive film material is shown in
From the response signals obtained from the MSS, 48-dimensional features were extracted. Then endpoint detection was performed, and three primary odors were selected from the prepared odor data. As the endpoint detection method, the example described in [Endpoint detection method for selecting primary odors] was used. The parameter used in the endpoint detection is set to K=10,000, and the total endpoint score is 20,000. The ranking of the endpoint scores is shown in the table below. The top three samples are fish sauce, Japanese rice wine, and pure water. Half or more of the total endpoint score is occupied by the top two odors, and it is considered that fish sauce and Japanese rice wine are reliably positioned at the endpoints among the prepared pure water and seasonings. This is consistent with the fact confirmed from the raw response signals. On the other hand, the endpoint score of the third-ranked pure water is a relatively small value as compared with these two odors. In such a case, it is difficult to directly find this rank by visual observation, simple comparison between feature vectors, or the like. The present invention is characterized in that it is possible to cope with such a case by performing endpoint detection. Furthermore, since the endpoint scores of grilled meat sauce, oyster sauce, and Worcestershire sauce are considerably small in the data of examples, it can be seen that these odors are located at a deeper area inside the region where the feature vectors are distributed in the 48-dimensional feature space. In this way, by executing the endpoint detection, the relative position in the feature space among the prepared odors can be find out. From the result of the endpoint detection, the primary odors can be determined to be fish sauce, Japanese rice wine, and pure water in the prepared data set of pure water and seasonings.
In order to convert the target odor into color, the coefficients (w1, w2, w3) are calculated by quadratic programming. (Details are stated in [Method of representing any odor using primary odors] described above.) These coefficients represent the linear combination of the standardized feature vectors
x
e
,x
e
, and xe
of the three primary odors extracted from the response signals. As the three primary odors, fish sauce, Japanese rice wine, and pure water are used as selected above. The following table shows the results of determining these coefficients by quadratic programming. The difference A between the linear combination and the standardized feature vector of the target odor is also listed in the table.
The value of Δ is the accuracy when the odor sample to be targeted is represented by feature vectors of three primary odors. This means that a sample having a small value of Δ, such as Worcestershire sauce or grilled meat sauce, can be represented relatively accurately by mixing the three primary odors. On the other hand, it has been confirmed that the mean value of Δ is increased when odors other than those of fish sauce, Japanese rice wine, and pure water are used as the primary odors. This fact means that the endpoint detection works effectively for achievement of better color representation. Further, by using these coefficients, the color for each odor sample, the position of each color in the chromaticity triangle, and the RGB value of each odor are summarized in
This color map indicates that the odors of soy sauce and Japanese noodle sauce do not contain a component of pure water and these odors are represented as a mixture of fish sauce and Japanese rice wine. It should be noted that the color itself has no meaning in the above expression. This is because colors unrelated to the original odor sample were assigned to the three primary odors, and thus the other odor samples were converted to colors by relative evaluation using these colors. On the other hand, it is also possible to give meaning to the color itself by applying the color of the liquid sample itself or a color tailored to human senses to the three primary odors using the prepared data set and by creating a chromaticity triangle by mixing such colors.
In
When three kinds of primary odors are determined in advance, various kinds of odors can be immediately converted into a form that can be recognized by perception other than olfaction, such as color, only by executing quadratic programming. Therefore, for example, the color for the target odor (color converted from odor as described above) can be output in real time. By utilizing such a technique, a device for displaying the color for odor in real time was produced using the MSS, a compact computer as an information processing device that implements machine learning, and an LED light. The LED light was used as the device for outputting the color, but of course, any output device may be employed. For three kinds of primary odors, fish sauce, Japanese rice wine, and pure water were used as an example. Further, the colors for the primary odors were set to red, green, and blue. Since the N2 gas was used as the reference gas and the carrier gas at the time of measuring the response signals of the primary odors, the N2 gas was used as the reference gas and the carrier gas also at the time of measuring the target odors. However, air can also be used as reference gas and carrier gas.
The real-time measurement results are shown in
[Details of Examples—Production of MSS Chip Coated with Various Materials]
A total of 12 channels (hereinafter, also referred to as Ch) are mounted on the MSS chips used in the present examples, and the material applied to each channel and the coating condition are as follows.
Ch1: Aminopropyl group-modified silica/titania composite nanoparticles (spray coating);
Ch2: Octadecyl group-modified silica/titania composite nanoparticles (inkjet coating, 1 g/L, 400 shots);
Ch3: Phenyl group-modified silica/titania composite nanoparticles (inkjet coating, 1 g/L, 200 shots);
Ch4: Polymethyl methacrylate (inkjet coating, 1 g/L, 300 shots); Ch5: Aminopropyl group-modified silica/titania composite nanoparticles (inkjet coating, 1 g/L, 600 shots);
Ch6: Octadecyl group-modified silica/titania composite nanoparticles (inkjet coating, 1 g/L, 800 shots);
Ch7: Phenyl group-modified silica/titania composite nanoparticles (inkjet coating, 1 g/L, 500 shots);
Ch8: Silica/hexadecyltrimethylammonium composite particles (inkjet coating, 1 g/L, 1000 shots);
Ch9: Tenax TA (mesh: 20/35) (inkjet coating, 1 g/L, 300 shots, stage temperature: 100° C.);
Ch10: Tenax TA (mesh: 20/35) (inkjet coating, 1 g/L, 300 shots, stage temperature: 20° C.);
Ch11: Tenax TA (mesh: 60/80) (inkjet coating, 1 g/L, 300 shots, stage temperature: 100° C.); and
Ch12: Tenax TA (mesh: 60/80) (inkjet coating, 1 g/L, 300 shots, stage temperature: 20° C.).
Various silica/titania composite nanoparticles applied to Ch1, Ch2, Ch3, Ch5, Ch6, and Ch7 were synthesized in accordance with methods that are well known to those skilled in the art and described in Non Patent Literature 6 and Non Patent Literature 7. The silica/hexadecyltrimethylammonium composite particles applied to Ch8 were synthesized in accordance with a method that is also well known to those skilled in the art and described in Non Patent Literature 8. The materials applied to Ch4, Ch9, Ch10, Ch11, and Ch12 were all commercially available, and the purchased materials were used as they were without purification or the like.
Various materials were each applied onto the MSS chip as follows. Ch1 was coated by spraying, and the others were coated by using ink jet. These coating methods are well known to those skilled in the art, but if needed, see Non Patent Literature 7 for details of spray coating and Non Patent Literature 9 for details of inkjet coating. It should be noted that, although any of the materials Ch2 to Ch12 was used for coating by preparing a dispersion liquid or a solution having a concentration of 1 g/L, the conditions such as the number of shots and the stage temperature are different for each channel.
The field of application of the present invention is not limited to odor, and the present invention can be used in any fields as long as it is useful in such a field to select original samples, that is, primary odors, from all kinds of gas, liquid, and solid samples. For example, by using a relatively small number of primary odors, any odor can be represented, presented, or synthesized. In addition, an example of application can be conceived as follows. Original samples (primary odors) are selected from various samples such as exhaled breath, sweat, saliva, tears, and any other body fluid, gas or odor emitted from the body. Then, the odor of any sample to be measured, that is, the odor of any individual, can be represented as a combination of the primary odors selected in this manner. Further, this combination can be represented in a form that can be recognized by other perceptions, for example, color. In addition, the present invention is expected to be widely used in various fields where odors or gases are generated, such as production of food, product design, entertainment, storage, distribution, security, and pharmaceutical fields.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2020-051050 | Mar 2020 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2021/008633 | 3/5/2021 | WO |
