Component odor determining apparatus

Abstract

To provide an element odor determination device improved in reproduction accuracy of an odor by considering human perception.SOLUTION: The component odor determining apparatus comprises an analyzing part 10, an artificial intelligence part 20, a component odor recipe determining part 30, and a blended odor recipe determining part 40. The artificial intelligence unit 20 maps the odor information to the olfactory information. A mixed odor recipe determination part 40 compares olfactory information of a temporary mixed odor generated by mixing a plurality of temporary element odors generated by mixing a plurality of single odors at a predetermined composition ratio at a predetermined mixing ratio with olfactory information of an object odor while updating the mixing ratio of the temporary element odors, and determines the most approximate mixing ratio. An element odor recipe determination part 30 evaluates the olfactory information of the mixed odor mixed by the mixing ratio determined by the mixed odor recipe determination part 40 while updating the constitution ratio of the single odor, and determines the constitution ratio of the plurality of single odors for respectively generating the most appropriate element odor.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an element odor determining device, and more particularly to an element odor determining device for determining element odors used in an olfactory display in which a target odor is produced by blending a plurality of element odors. [Background technology]

[0002] While the three primary colors that reproduce colors are well known, the original odors that reproduce scents have not yet been identified. Therefore, with regard to olfaction, it is not common to reproduce original odors based on a set of principles. However, in recent years, olfactory displays that generate target odors by blending multiple elemental odors have been developed. Olfactory displays are devices that provide various odors selected by the user by blending elemental odors selected from many elemental odors. Olfactory displays require the use of a large number of suitable elemental odors in order to provide a variety of odors that are closer to the actual odors.

[0003] Here, component odors are generated by mixing multiple single odors at a predetermined composition ratio. A known method for determining component odors is disclosed in Patent Document 1, which was submitted by the same applicant as the present application. In Patent Document 1, the mass spectrum of a sample odor is measured using a mass spectrometer, and basis vectors are extracted from the data using the NMF (Non-negative Matrix Factorization) method. The composition ratio is then determined so that the mass spectrum of the odor generated by mixing multiple single odors at a predetermined composition ratio matches the basis vector. Thus, the odor generated by mixing multiple single odors at that composition ratio is considered to be the component odor. In other words, the composition ratio of the multiple single odors that make up the component odor is determined by comparing their mass spectra. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2009-300188 Summary of the Invention [Problem to be solved by the invention]

[0005] As in Patent Document 1, the composition ratio of a plurality of single odors that make up the elemental odors may be determined using only mass spectra, but further improvement in the accuracy of reproducing the aroma has been desired.

[0006] Here, if the composition ratio of the single odors that make up the elemental odors could be determined using human perception, such as sweet or sour odors, the accuracy of scent reproduction could be improved. However, it is not realistic in terms of frequency and time for a human to sniff the odors every time when determining the composition ratio of the multiple single odors that make up the elemental odors or when determining the blended odor of the elemental odors that make up the target odor. Furthermore, if the senses become numb from sniffing the odors multiple times, the accuracy could deteriorate.

[0007] In view of the above circumstances, the present invention aims to provide an element odor determining device that improves the accuracy of reproducing aromas by taking into account human perception. [Means for solving the problem]

[0008] In order to achieve the above-mentioned object of the present invention, the component odor determination device of the present invention comprises an analysis unit that analyzes odors and generates odor information; an artificial intelligence unit that maps the input odor information to olfactory information; a blended odor recipe determination unit that compares olfactory information mapped using the analysis unit and the artificial intelligence unit for a provisional blended odor generated by blending a plurality of provisional component odors, each generated by mixing a plurality of simple odors at a predetermined composition ratio, with olfactory information of a target odor while updating the composition ratio of the provisional component odors, and determines the most approximate composition ratio; and a component odor recipe determination unit that evaluates the olfactory information mapped using the analysis unit and the artificial intelligence unit for the blended odor blended at the composition ratio determined by the blended odor recipe determination unit while updating the composition ratio of the simple odors, and determines the composition ratio of the simple odors for generating the most appropriate component odors.

[0009] Here, the blended odor recipe determination unit may include a comparison unit that evaluates the olfactory information of the provisional blended odor and the target odor using a cost function.

[0010] Furthermore, the component odor recipe determination unit may include an appropriateness evaluation unit that evaluates, by a cost function, olfactory information mapped to the blended odor using the analysis unit and the artificial intelligence unit.

[0011] Furthermore, the cost function may be composed of a term for minimizing errors in olfactory information, a term for making the sum of the component ratios of multiple single odors approach 100%, and a term for making the vectors of each virtual element odor as orthogonal as possible.

[0012] Furthermore, the component odor recipe determination unit may use the line search method or the gradient method when updating the component ratios of multiple single odors.

[0013] Furthermore, the blended odor recipe determination unit may use the line search method or the gradient method when updating the blend ratio of the provisional element odors.

[0014] The analysis unit may be any of a mass spectrometer, an IMS (Ion Mobility Spectrometry) device, a gas chromatograph, an odor sensor, and an odor biosensor that can generate a mass spectrum as odor information.

[0015] Furthermore, the artificial intelligence unit may be one that uses a deep learning neural network.

[0016] Furthermore, the artificial intelligence unit may be configured to learn using a plurality of single odors individually.

[0017] Furthermore, the olfactory information of the artificial intelligence unit may be a plurality of scent descriptors or a proportion belonging to a plurality of scent descriptors.

[0018] The artificial intelligence unit may also improve prediction accuracy by clustering the similarities between scent descriptors using natural language processing.

[0019] The olfactory information of the artificial intelligence unit may be either receptor response or human perception.

[0020] The element odor determining device of the present invention has the advantage that the accuracy of reproducing the scent is improved by taking human perception into consideration. [Brief explanation of the drawings]

[0021] [Figure 1] FIG. 1 is a schematic block diagram for explaining the component odor determining device of the present invention. [Figure 2] FIG. 2 is a block diagram for explaining the blended odor recipe determination unit of the component odor determination device of the present invention. [Figure 3] FIG. 3 is a diagram for explaining the interrelationships between scent descriptors used in the element odor determination device of the present invention. [Figure 4] FIG. 4 is a graph illustrating the effect of clustering scent descriptors, which are olfactory information, in the element odor determination device of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The component odor determination device of the present invention is for determining component odors to be used in an olfactory display that generates a target odor by blending multiple component odors. The component odor determination device of the present invention first mixes multiple single odors in a predetermined composition ratio to generate multiple provisional component odors, and then optimizes the provisional blended odor generated by blending these provisional component odors in a predetermined composition ratio. The component ratios of the multiple single odors used to generate each of the component odors that make up the optimized blended odor are then optimized. This makes it possible to determine the component ratios of the single odors that make up the optimal component odors. Here, the single odor may be a single compound or a composite odor obtained by blending multiple compounds.

[0023] 1 is a schematic block diagram for explaining the component odor determination device of the present invention. As shown in the figure, the component odor determination device of the present invention is mainly composed of an analysis unit 10, an artificial intelligence unit 20, and a component odor recipe determination unit 30.

[0024] The analysis unit 10 analyzes odors and generates odor information. The analysis unit 10 may be any unit capable of analyzing an input odor and outputting predetermined analytical information that can identify the odor. Specifically, the analysis unit 10 may be any unit capable of generating, for example, a mass spectrum as odor information. Here, a mass spectrum is a two-dimensional chart in which the horizontal axis represents m / z (m is the mass of the ion divided by the unified electron mass unit, and Z is the charge of the ion) and the vertical axis represents the detection intensity. Because it shows a pattern specific to a substance, it is possible to estimate the chemical composition and molecular structure of the input odor. Specific examples of the analysis unit 10 include a mass spectrometer, an IMS (Ion Mobility Spectrometry) device, a gas chromatography device, an odor sensor, an odor biosensor, and the like. The analysis unit 10 may be any existing or future device capable of analyzing odors and generating odor information.

[0025] The artificial intelligence unit 20 maps the input odor information into olfactory information. Here, the olfactory information may be, for example, multiple scent descriptors. Scent descriptors are sensory expressions of human perception, such as sweet, sour, or refreshing. That is, the artificial intelligence unit 20 converts the mass spectrum, which is odor information, into a sensory expression of human perception. The mapped olfactory information may be a ratio belonging to multiple scent descriptors. That is, it may be a sensory expression combining multiple scent descriptors in a predetermined ratio. Furthermore, the olfactory information may not only be human perception but also a receptor response. A receptor response is, for example, a response reaction of olfactory chemosensory receptors, and this may be used as sensory information. Specifically, the artificial intelligence unit 20 can use a deep learning neural network. The artificial intelligence unit 20 can learn by using each of the multiple single odors that make up the element odors individually. This significantly reduces the amount of data required for the vast number of combinations of mixed odors, enabling training data to be acquired with a reasonable amount of effort. The artificial intelligence unit 20 can be any existing or future developed unit that can map odor information into olfactory information.

[0026] Next, we will explain the component odor recipe determination unit 30, which is a characteristic part of the component odor determination device of the present invention. First, the component odor recipe determination unit 30 determines the appropriate composition ratio of multiple single odors for generating component odors using the above-mentioned analysis unit 10 and artificial intelligence unit 20. The component odor recipe determination unit 30 is composed of a provisional composition ratio determination unit 31, a blended odor recipe determination unit 40, a blended odor generation unit 32, an appropriateness evaluation unit 33, a single odor composition ratio update unit 34, and a component odor generation unit 35.

[0027] First, the provisional component ratio determination unit 31 determines in advance a plurality of provisional component odors to be generated by mixing a plurality of tentative component odors at a predetermined component ratio. That is, the provisional component ratio determination unit 31 stores an initial value for each tentative component odor. The provisional component ratios of the tentative component odors may be determined arbitrarily in advance. The blended odor recipe determination unit 40 then optimizes the blend ratio of the provisional blended odor generated by blending a plurality of tentative component odors at a predetermined blend ratio. The blended odor generation unit 32 generates a blended odor using the optimized blend ratio. The appropriateness evaluation unit 33 evaluates the olfactory information mapped to the generated blended odor using the analysis unit 10 and the artificial intelligence unit 20. If the appropriateness evaluation unit 33 determines that the blended odor is inappropriate, the single odor component ratio update unit 34 updates the component ratios of the single odors. The component odor generation unit 35 generates each provisional component odor based on the updated component ratios and provides feedback. If the feedback ultimately determines that the component ratios are appropriate, the current component ratios are determined to be appropriate component ratios of the multiple single odors for composing a component odor with appropriate olfactory information.

[0028] Specifically, as shown in Fig. 1, the component odor recipe determination unit 30 has a feedback loop consisting of an appropriateness evaluation unit 33, a single odor composition ratio update unit 34, a component odor generation unit 35, a blended odor recipe determination unit 40, and a blended odor generation unit 32. First, the tentative composition ratio determination unit 31 determines, for example, eight tentative component odors (tentative component odors 1, 2, 3...8). The tentative component odors are generated by mixing multiple single odors at a predetermined composition ratio. That is, the tentative component odors are configured as follows: Temporary element odor 1=C 11 :C 12 :C 13 ... Pseudo-element odor 2 = C 21 :C 22 :C 23 ... : Temporary element odor 8=C 81 :C 82 :C 83 ...

[0029] Then, the blended odor recipe determination unit 40 determines the composition ratio of, for example, 96 provisional blended odors at a predetermined blend ratio using these provisional element odors 1-8. That is, the provisional blended odors are configured as follows. Provisional blended odor 1 = Provisional element odor 1: Provisional element odor 2: Provisional element odor 8 Provisional blended odor 2 = Provisional element odor 1: Provisional element odor 2: ··· Provisional element odor 8 : Provisional blended odor 96 = Provisional element odor 1: Provisional element odor 2: Provisional element odor 8

[0030] The appropriateness evaluation unit 33 evaluates whether the olfactory information of the blended odor generated by the blended odor generation unit 32 is appropriate. The single odor component ratio update unit 34 changes the component ratio of the single odors for the tentative component odor 1 of the multiple tentative component odors 1-8 that make up the blended odor, and the appropriateness evaluation unit 33 evaluates whether the tentative component odor 1 is appropriate using a feedback loop. Once tentative component odor 1 is optimized, the appropriateness evaluation unit 33 evaluates whether the next tentative component odor 2 is appropriate using a feedback loop and optimizes it. This process is repeated up to tentative component odor 8. Then, returning to tentative component odor 1, the appropriateness evaluation unit 33 evaluates whether the next tentative component odor 2 is appropriate using a feedback loop and optimizes it. This process is repeated up to tentative component odor 8, and continues until convergence. As a result, the tentative component odors 1-8 at the final convergence become the component odors 1-8. Therefore, the component ratios of the multiple single odors that make up each of the component odors 1-8 at this time are determined. In this way, each component odor can be determined using the component ratios of the multiple single odors optimized using the blended odor determination device of the present invention. In an olfactory display, a target odor can be generated by blending these optimized component odors.

[0031] The blended odor recipe determination unit 40 will be described in detail using FIG. 2. FIG. 2 is a block diagram illustrating the blended odor recipe determination unit of the component odor determination device of the present invention. In the figure, parts with the same reference numerals as in FIG. 1 represent the same components. The blended odor recipe determination unit 40 has a feedback loop consisting of a comparison unit 41, a blending ratio update unit 42, and a blending unit 43. The blended odor recipe determination unit 40 determines the blending ratio of multiple component odors to generate a blended odor that will provide olfactory information similar to the target odor, using the analysis unit 10 and the artificial intelligence unit 20 described above. That is, the blended odor recipe determination unit 40 determines the blending ratio of multiple component odors to generate the target odor. First, the blending unit 43 blends multiple provisional component odors generated based on the component ratios determined by the provisional component ratio determination unit 31 and the component odor generation unit 35 at a predetermined blending ratio to generate a provisional blended odor. Then, the comparison unit 41 compares the olfactory information mapped to the provisional blended odor with the olfactory information of the target odor. Here, the olfactory information of the target odor may be the human perception or receptor response when the target odor is actually smelled. For example, it may be the sensory description of the sensation felt when 96 target odors are actually smelled by a human. If the comparison unit 41 determines that the odors are not similar, the blending ratio update unit 42 updates the blending ratio of the multiple provisional component odors. The blending unit 43 generates a new provisional blended odor based on the updated blending ratio and provides feedback. If the feedback ultimately determines that the odors are similar, the blending ratio at that time is determined to be the blending ratio of the multiple provisional component odors for generating a blended odor that will provide olfactory information similar to the target odor.

[0032] 1 and 2, the analysis unit 10 and the artificial intelligence unit 20 are shown as the same unit. The analysis unit 10 and the artificial intelligence unit 20 may be switched and connected to the component odor recipe determination unit 30 or the blended odor recipe determination unit 40 as appropriate. However, the present invention is not limited to this, and the analysis unit 10 and the artificial intelligence unit 20 may be provided in each of the component odor recipe determination unit 30 and the blended odor recipe determination unit 40. Furthermore, after multiple component odors are determined by the component odor recipe determination unit 30, the recipe for those component odors can be used. Therefore, in the operational stage of determining the blending ratio of the blended odor, the number of target odors and provisional blended odors is set to 1, and only the blended odor recipe determination unit 40, the analysis unit 10, and the artificial intelligence unit 20 are required.

[0033] In this way, the component odor determination device of the present invention uses human perception, such as sweet, sour, or refreshing, to determine the composition ratio of multiple single odors to compose component odors or the blend ratio of multiple component odors to create a blended odor, thereby improving the accuracy of fragrance reproduction. Human perception can be more sensitive than analytical instruments, which is advantageous for accurately reproducing complex sensations. Although human perception is taken into account when determining the component odors and blended odors, the blended odor determined by the component odor determination device of the present invention does not require a human to sniff the odors each time. Therefore, there is no need to consider the number of times or duration of sniffing, and the accuracy is realistic.

[0034] In the component odor determination device of the present invention, the artificial intelligence unit 20, the component odor recipe determination unit 30, and the blended odor recipe determination unit 40 can also be configured as a computer program or the like.

[0035] Here, the appropriateness evaluation unit 33 of the component odor recipe determination unit 30 will be specifically described. The appropriateness evaluation unit 33 of the component odor recipe determination unit 30 uses a cost function to evaluate the olfactory information mapped to the blended odor using the analysis unit 10 and the artificial intelligence unit 20. Here, the cost function is a function that represents the difference between the prediction by the artificial intelligence unit 20 and the original correct answer. In other words, the cost function adjusts the component ratio of multiple single odors that make up the component odor so as to minimize the error of the artificial intelligence unit 20. The appropriateness evaluation unit 33 calculates the difference between the olfactory information output as a result of learning by the artificial intelligence unit 20 and the actual olfactory information for the hypothetical component odor. If there is a difference, the component ratio of the multiple single odors that make up the component odor is updated by the single odor component ratio update unit 34. Then, feedback is performed until the difference is minimized, and the final appropriate component ratio is determined.

[0036] More specifically, the cost function should consist of a term for minimizing errors in olfactory information, a term for making the sum of the component ratios of multiple single odors approach 100%, and a term for making the vectors of each virtual element odor as orthogonal as possible.

[0037] First, the first term to minimize the error in olfactory information is given by the following formula:

number

[0038] Next, the second term, which ensures that the sum of the component ratios of multiple single odors approaches 100%, is given by the following formula:

number

[0039] The third term, which ensures that the vectors of each pseudo-element odor are as orthogonal as possible, is given by the following formula:

number

[0040] The cost function is the weighted sum of the three terms defined in this way. Regarding the importance of different parts, the weights of the three terms are set to, for example, 1, 100, and 300, respectively. These weights can be determined so that each term works appropriately.

[0041] The appropriateness evaluation unit 33 can use such a cost function to determine the composition ratio of multiple single odors to generate the most appropriate element odor that improves the cost function.

[0042] The comparison unit 41 of the blended odor recipe determination unit 40 can also evaluate the olfactory information of the provisional blended odor and the target odor using a cost function similar to that described above. That is, in the above-mentioned cost function, N=1 is used to compare one target odor with the provisional blended odor, and the most approximate blending ratio that improves the cost function can be determined.

[0043] Furthermore, if the appropriateness evaluation unit 33 determines that the odor is inappropriate, the component ratios of the multiple single odors that make up the component odors are updated by the single odor component ratio update unit 34. Here, the single odor component ratio update unit 34 may use the line search method or gradient method when updating the component ratios of each tentative component odor until the olfactory information becomes appropriate. The component ratios of the multiple single odors that make up the component odors are updated using the line search method or gradient method, and are fed back to the appropriateness evaluation unit 33 so that the difference is minimized.

[0044] Furthermore, if the olfactory information input to the comparison unit 41 of the blended odor recipe determination unit 40 differs from the olfactory information mapped to the target odor, the blending ratio update unit 42 updates the provisional blending ratio of the multiple component odors for generating the blended odor. Here, the blending ratio update unit 42 may also use the line search method or gradient method when updating the blending ratio of the provisional blended odor until the olfactory information matches. The blending ratio of the multiple component odors for generating the blended odor is updated using the line search method or gradient method, and is fed back to the comparison unit 41 so that the blended odor matches.

[0045] Furthermore, the artificial intelligence unit 20 can improve prediction accuracy by clustering the similarities between scent descriptors, which are olfactory information, using natural language processing. The details of clustering are explained below. Figure 3 is a diagram illustrating the interrelationships between scent descriptors used in the element odor determination device of the present invention. Specifically, Figure 3 shows the similarities between scent descriptors using hierarchical clustering. Using a natural language processing method (here, FastText), Wikipedia was trained as a corpus and each descriptor was represented as a 300-dimensional vector. Cluster analysis was then performed using the cosine distance between the 300-dimensional vectors as a measure of similarity. For example, some words, such as "oily" and "fatty," are close to each other and consistent with intuition. More specifically, the cutoff line is set to 0.031, and a cluster label is assigned to each scent descriptor. This allows, for example, 37 scent descriptors to be clustered into 13. In other words, the original 37 scent descriptors can be rewritten using 13 cluster labels. By performing clustering in this manner, the artificial intelligence unit 20 improves prediction accuracy.

[0046] FIG. 4 is a graph illustrating the effect of clustering scent descriptors, which are olfactory information, in the elemental odor determination device of the present invention as described above. Specifically, FIG. 4 shows the precision, recall, and balanced accuracy of the reproduced scent descriptors before and after clustering. Here, the number of elemental odors was eight. Furthermore, 96 essential oil scents were used as samples, and a mass spectrometer was used as the analysis unit 10. As shown in the figure, by clustering the similarities between scent descriptors using natural language processing, the precision improved from 0.5 to 0.6, the recall improved from 0.4 to 0.6, and the balanced accuracy improved from 0.67 to 0.74. Therefore, it can be seen that clustering scent descriptors, which are olfactory information, in the artificial intelligence unit 20 can further improve the accuracy of scent reproduction.

[0047] The component odor determination device of the present invention is not limited to the above-described illustrated example, and it goes without saying that various modifications can be made without departing from the spirit of the present invention. [Explanation of symbols]

[0048] 10 Analysis Department 20 Artificial Intelligence Department 30 Elemental Odor Recipe Determination Section 31 Provisional Composition Ratio Determination Department 32 Blended odor generating section 33 Appropriateness Assessment Department 34 Single Odor Composition Ratio Update Section 35 Elemental odor generator 40 Blended Odor Recipe Decision Department 41 Comparison section 42 Mixing ratio update section 43 Mixing Department

Claims

1. An element odor determining device for determining element odors to be used in an olfactory display that generates a target odor by blending a plurality of element odors, the element odor determining device comprising: an analysis unit that analyzes odors and generates odor information; an artificial intelligence unit that maps input odor information into olfactory information; a blended odor recipe determination unit that compares olfactory information of a target odor mapped by the analysis unit and the artificial intelligence unit with olfactory information of a provisional blended odor generated by blending a plurality of provisional element odors, each of which is generated by mixing a plurality of single odors at a predetermined composition ratio, while updating the blending ratio of the provisional element odors, and determines the most approximate blending ratio; an element odor recipe determination unit that evaluates olfactory information mapped by the analysis unit and the artificial intelligence unit to the blended odors blended at the blending ratio determined by the blended odor recipe determination unit while updating the component ratios of the single odors, and determines the component ratios of the multiple single odors for generating the most appropriate element odors; A component odor determining device comprising:

2. 2. The component odor determination device according to claim 1, wherein the blended odor recipe determination unit has a comparison unit that evaluates olfactory information of the provisional blended odor and the target odor using a cost function.

3. 2. The component odor determination device according to claim 1, wherein the component odor recipe determination unit has an appropriateness evaluation unit that evaluates olfactory information mapped to a blended odor using the analysis unit and the artificial intelligence unit using a cost function.

4. In the component odor determination device described in claim 2 or claim 3, the cost function is characterized in that it is composed of a term for minimizing errors in olfactory information, a term for making the sum of the component ratios of multiple single odors approach 100%, and a term for making the vectors of each virtual component odor as orthogonal as possible.

5. 2. The component odor determination device according to claim 1, wherein the component odor recipe determination unit uses a line search method or a gradient method when updating the component ratios of a plurality of single odors.

6. 2. The component odor determination device according to claim 1, wherein the blended odor recipe determination unit uses a line search method or a gradient method when updating the blend ratio of the tentative component odors.

7. 2. The component odor determination device according to claim 1, wherein the analysis unit uses any one of a mass spectrometer, an IMS (Ion Mobility Spectrometry) device, a gas chromatography device, an odor sensor, and an odor biosensor, all of which are capable of generating a mass spectrum as odor information.

8. 2. The component odor determination device according to claim 1, wherein the artificial intelligence unit uses a deep learning neural network.

9. 9. The component odor determination device according to claim 8, wherein the artificial intelligence unit learns each of a plurality of single odors individually.

10. 2. The element odor determination device according to claim 1, wherein the olfactory information of the artificial intelligence unit is a plurality of scent descriptors or a proportion belonging to a plurality of scent descriptors.

11. 11. The element odor determination device according to claim 10, wherein the artificial intelligence unit improves prediction accuracy by clustering the similarities between scent descriptors using natural language processing.

12. 2. The element odor determining device according to claim 1, wherein the olfactory information of said artificial intelligence unit is either receptor response or human perception.

Images