JPH06174689A - Method of extracting characteristic of taste - Google Patents

Abstract

PURPOSE:To easily grasp the difference of tastes by performing a principal component analysis of much data which is obtained by measuring a plurality of solutions to be measured by means of a taste sensor having a plurality of fat membranes. CONSTITUTION:Substantial tastes contributing to the difference of tastes among identical foods are two or three. Data related to tastes of a plurality of solutions to be measured 11- are measured by utilizing a multichannel type taste sensor having a plurality of fat membranes (14-1)-(14-8) by each channel. An electric signal from each fat membrane 14 is operated by a microcomputer 22. The microcomputer performs a principal component analysis of the data, that is, many values of related variables are indicated with a few synthetic variables to determine components which form first, second and third axes. On a plane determined by the first, second and third axes, solutions 11- are defined on the basis of the data converted in the principal component analysis process. Thereby, it is possible to easily grasp the difference of tastes of each solution.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention utilizes a sensor capable of substituting one of the five senses of human beings, the difference in taste of food and drink, which has been difficult to measure with an artificial sensor. The present invention relates to a technology that enables detection and measurement of. Food such as drinking water to be eaten and drink, the difference in taste such as alcoholic beverages, because it is to provide a technique for detecting what should be said the difference in taste, in the production plant of drinking water and liquor,
The present invention relates to a technology that enables quality control to be performed by a mechanical device without manual labor.

[0002]

[Meaning of terms] It is said that there are salt, sweetness, bitterness, sourness, and umami as basic elements of taste, and it is said that each has a different degree. These taste differences that can be evaluated by the human sense, or saltiness (similar kind) differences in saltiness, can be grasped as a physically measurable amount, and the measurable taste or taste difference. (Comparative or contrasting taste) is referred to herein as "horse mackerel".

[0003]

2. Description of the Related Art Conventionally, for example, as disclosed in Japanese Unexamined Patent Publication No. 62-187252, the original taste (basic taste) component in a measurement object, that is, the concentration of a selected taste substance is determined from the output values of a plurality of taste sensors. The horse mackerel was measured by calculating and correcting each concentration value to a value representing the strength of each original taste that matches the taste of a person. However, the taste sensor referred to in the above publication is a chemical sensor or a physical sensor that selectively detects a substance exhibiting each basic taste. Specifically, saltiness is a salt concentration meter, acidity is a hydrogen ion index meter, and sweetness is sweetness. Was a sugar meter utilizing the refractive index of the liquid to be measured. Since these sensors are selective, for example, a salt concentration meter that is trying to measure the strength of saltiness can measure the concentration of salt, but it cannot measure the concentration of other substances that have a salty taste and is suitable for human taste. There was a limit to how to correct it. If you compare this to color,
It was like trying to obtain color results using a sensor that only detects a single color.

The applicant of the present application, together with the joint applicant, has already filed a patent application for "taste sensor and its manufacturing method" (Japanese Patent Application No. 1-190819). In the specification and drawings of this application, a lipid substance composed of a molecule having a hydrophobic portion and a hydrophilic portion is fixed in a polymer matrix, and the hydrophilic portion of the lipid molecule is attached to the surface of the substance. The lipidic molecular film with a structure that aligns the
It has been shown that it can serve as a taste sensor that can replace the human taste.

FIG. 4 is a diagram showing the membrane formula of the lipidic molecular membrane by the expression method used in the method of designing chemical substances. The spherical portion indicated by a circle in the lipidic molecule is a hydrophilic group a, that is, a hydrophilic portion a, and has a hydrocarbon chain structure b (for example, an alkyl group) from which the atomic arrangement extends long. In each figure, two chains extend in each case to represent one molecule, and the whole constitutes a molecule group. The portion of this hydrocarbon chain is the hydrophobic site b. Such a lipid molecule group 31 is partly dissolved in the matrix 33 (surface structure, microscopic structure having a planar spread) on the surface of the membrane member 32 (eg It is accommodated at 31 ') in FIG. The way it is stored is
The hydrophilic parts are arranged on the surface.

FIGS. 5 (a) and 5 (b) show a multi-channel taste sensor using this lipidic molecular film. In this figure, three sensitive parts of the multi-channel array electrode are shown. In the illustrated example, a 0.5 mmφ hole was penetrated through the base material, and a silver rod was inserted into the hole to form an electrode. The lipidic molecular film is attached to the substrate so as to contact the electrode via the buffer layer.

FIG. 6 shows a horse mackerel measuring system using the multi-channel taste sensor. Make an aqueous solution of the taste substance, use it as the solution to be measured 11, and put it in a container 12 such as a beaker.
Put in. The taste sensor array 13 prepared by arranging the lipid film and the electrode on the acrylic plate (base material) as described above was put in the solution to be measured. 1m potassium chloride before use
The electrode potential was stabilized with a mole / l aqueous solution. 14- in the figure
1, ... 14-8 are the lipid membranes shown by black dots. A reference electrode 15 is prepared as an electrode that generates an electric potential that serves as a reference for measurement, and the reference electrode 15 is placed in the solution to be measured. The taste sensor array 13 and the reference electrode 15 are installed at a predetermined distance. The surface of the reference electrode 15 is covered with 100 m mole / l of potassium chloride solidified with agar as the buffer layer 16, so that the electrode system is eventually silver 2 | silver chloride 4 | lipid membrane 3 (14) |
Solution to be measured 12 ｜ Buffer layer (potassium chloride 100m mole / l)
16 | Silver chloride 4 | Silver 2

The electric signal from the lipid membrane becomes a signal of 8 channels in the figure and is led to the buffer amplifiers 19-1, ..., 19-8 by the lead wires 17-1 ,. Each output of the buffer amplifier 19 is selected by the analog switch (8 channels) 20 and added to the A / D converter 21. The electric signal from the reference electrode 15 is also applied to the A / D converter 21 via the lead wire 18, and the difference from the potential from the membrane is converted into a digital signal. This digital signal is appropriately processed by the microcomputer 22 and displayed by the XY recorder 23. In this example, an 8-channel taste sensor is used, and each channel has different response characteristics to taste in order to obtain a large amount of taste information capable of reproducing the taste of human being. It is composed of a molecular film.

[0009]

[Table 1]

The applicant of the present application, together with the joint applicant,
We also filed a patent application for "Taste sensor and its manufacturing method" (Japanese Patent Application No. 3-020450). In the specification and drawings of this application, a molecular film closer to the human taste organ than the previous application (Japanese Patent Application No. 1-190819) is shown. Then, the structure of a molecular film in which a substance called an amphipathic substance having a hydrophilic group and a hydrophobic group (including a lipid) or a bitter substance such as an alkaloid is used as a material of the molecular film is shown. As shown in FIG. 7, the structure is such that the amphipathic molecule group 36 or the bitter substance molecule group 36 is aligned on the base film 7 provided on the substrate 1 with the hydrophilic portions indicated by circles facing outward, It constitutes a monolayer.

The taste sensor referred to in the above specification is just a taste sensor, has physicochemical properties close to those of the tongue, which is the taste organ of humans, and has similar tastes even if the taste substances are different. You can get output, and you can get some output for different tastes. For example, this is a sensor that can detect color.

When the difference in the horse mackerel of each solution to be measured is grasped from the data obtained by these taste sensors, the data relating to the horse mackerel of each solution to be measured obtained by, for example, an 8-channel taste sensor is shown in FIG. As shown in the figure, the horizontal axis is the channel, the output value for each measured solution of each channel is plotted on the vertical axis, and the shape of the pattern obtained by connecting the plotted points for each measured solution shows the distance between each measured solution. The difference in horse mackerel between each solution to be measured can be understood from the shape of the pattern obtained by grasping the difference in horse mackerel or by displaying points on a radar chart as shown in Fig. 8 and plotting points for each solution to be measured. Was trying to figure out.

[0013]

In the conventional method of grasping horse mackerel, it is extremely difficult to intuitively grasp the distance of the horse mackerel of the solution to be measured from the data of, for example, 8 channels (that is, 8 dimensions). In addition, it is extremely difficult to intuitively grasp even if the distance and the direction are obtained by performing an eight-dimensional vector operation. Further, there is a problem that the correlation between the sensors is strong and the characteristics are buried, and it is difficult to grasp the difference in horse mackerel between the solutions to be measured.

The object of the present invention is to solve these problems and to provide a taste sensor using a lipid membrane (as described in the section of the prior art, the materials that can be used for the taste sensor are not only lipids but also amphipathic substances). It has spread to substances and bitter substances,
The lipid membrane will be explained as a representative example. ) Is used to extract the characteristics from the horse mackerel measurement data, and to provide a method for extracting the horse mackerel characteristics that makes it easy to grasp the difference in the horse mackerel between the solutions to be measured.

[0015]

[Means for Solving the Problems] As mentioned above, the taste is said to consist of five basic tastes. Of the same foods, there are often two or three basic tastes that contribute to the difference in horse mackerel. For example, in beer, bitterness and umami contribute to horse mackerel, and saltiness, sweetness, and sourness are extremely weaker than these.

On the other hand, in the taste sensor, the reaction with respect to horse mackerel differs depending on the lipid membrane used, and moreover, one kind of lipid film does not react with only one kind of horse mackerel. Therefore, we try to grasp the difference of horse mackerel from many data by using a multi-channel type taste sensor using multiple lipid membranes. However, if the data is compared as it is, the amount of data is large, it is difficult to intuitively grasp, and the characteristic of horse mackerel between samples may be buried.

Therefore, the method of locating each sample was decided by focusing on the component that best represents the difference in horse mackerel between each sample. That is, using a taste sensor having a plurality of channels using a plurality of types of lipid membrane to measure the data related to the aji of a plurality of measured solutions for each channel, and the plurality of measured solutions as a sample The step of determining the components to be the first axis, the second axis, and the third axis by performing the principal component analysis of the data using the plurality of channels as variables, and the first axis and the second axis or the third axis. Positioning the solution to be measured on a plane based on the data converted in the process of principal component analysis.

Further, since a lot of data obtained by the multi-channel type taste sensor also includes data that become noise, before taking correlation for each sample,
Correlation is performed for each channel, and a channel having a high contribution rate is selected, and the data of the channels up to a point where the total contribution rate exceeds 95% are used for obtaining the correlation for each sample. That is, using a taste sensor having a plurality of channels using a plurality of types of lipid membranes, measuring the data related to the azimuth of a plurality of solutions to be measured for each channel, and using the plurality of channels as a sample As a variable, the step of subjecting the data to principal component analysis, the step of selecting a desired plurality of channels from the plurality of channels from the result of the principal component analysis, and the plurality of subject solutions to be measured as samples. Using the selected plurality of channels as variables, principal component analysis of the data is performed to determine the first axis, the second axis, and the third axis.
The method comprises the steps of determining a component serving as an axis, and positioning the solution to be measured on a plane defined by the first axis and the second axis or the third axis based on the data converted in the process of principal component analysis. ing.

[0019]

In the feature extraction method for horse mackerel according to the first aspect of the present invention, principal component analysis is performed on data related to horse mackerel of a plurality of solutions to be measured obtained by using a taste sensor having a plurality of channels.
Based on the result, each solution to be measured is a biaxial plane or 3
It is located in the axial space. When the solutions to be measured are limited to foods of the same kind, the basic taste is two to three-dimensional, and thus the two or three axes can sufficiently express the difference in horse mackerel between the solutions to be measured. Moreover, two-dimensional and three-dimensional are easy for a person to grasp.

[0020]

EXAMPLE First, principal component analysis will be described. Principal component analysis is a method in which the values of many correlated variables are represented by decimal synthetic variables. At this time, this synthetic variable includes as much information as the original variable has, that is, it is an effective method of summarizing many variables and summarizing the phenomenon. The outline of the calculation process is shown below. The output of the sensor i (p ≧ i ≧ 1) is vi, and the output when the measured solution j is measured is vij. Here, the variable is the sensor and the sample is the solution to be measured. A composite variable Y obtained by linearly converting the sensor output (vi) from this data
Is

[0021]

[Equation 1]

However,

[0023]

[Equation 2]

It is assumed that The composite variable Y thus created must sufficiently reflect the outputs of p sensors. That is, the comprehensive index is to be found in the direction in which the scattering of each solution to be measured is the largest. This results in maximizing the variance of the composite variable Y. The variance V (Y) of the composite variable Y is

[0025]

[Equation 3]

However, a = (a1, a2, ..., ap)
Where aT is the transposed matrix of a, m is the number of samples, and S is the variance-covariance matrix due to the variable (sensor) of this data. Therefore, ai that maximizes Equation 3 is obtained under the constraint of Equation 1. This is the matrix S (variance covariance matrix)
Reduce to the eigenvalue problem of. The matrix S has p eigenvalues (λ1 ≧ λ2 ≧ ... ≧ λp), in which
The synthetic transformation Y1 that maximizes Equation 3 is such that the element of the eigenvector for the maximum eigenvalue λ1 is the coefficient a1i,

[0027]

[Equation 4]

Is represented by This synthetic variable Y1 is called the first principal component. The variance of this composite variable Y1 is λ1,
Largest variance. Therefore, it is a synthetic variate that includes the information of p pieces of variates (sensors) most. Using the elements of the eigenvector for λj as coefficients,

[0029]

[Equation 5]

It is assumed that This synthetic variable Yj is called the j-th principal component and has the j-th largest variance. An index of how much each principal component reflects the original data is the contribution rate Kj. In general, the variance of each principal component is represented by its eigenvalue λj, and its sum is equal to the sum of variances of the original variables.
Therefore,

[0031]

[Equation 6]

Defined as In addition, the total of the first to j-th contribution rates is called a cumulative contribution rate. Here, FIG. 8 shows a radar chart created from data obtained by the taste sensor of 16ch for 12 kinds of sake. Further, FIG. 9 shows the result of principal component analysis of the same data. here,
Think about the distinction between pure rice and brewed sake. Junmaishu is sake made only from rice, and brewed sake is sake made by adding brewing alcohol and the like. Junmaishu is 1, 2, 3,
7, 10 and 11 liquors, the other is brewed liquor. Pure rice sake is represented by black circles and brewed sake is represented by white circles. In the radar chart of FIG. 8, it is difficult to distinguish the difference between the two. However, in the principal component analysis, the brewed liquor group is centered in the center of the figure and is very easy to identify. As the number of samples increases, the state of the boundary line between the two becomes more apparent and the identification becomes more accurate. On the other hand, in the radar chart, as the number of samples increases, the type of each pattern increases, and it becomes more and more difficult to discriminate between the two from the shape of the pattern. Further, the more complicated the boundary of discrimination, the greater the effect of the principal component analysis. If we limit the number of samples to only liquor if it is liquor even if the number of samples is increased, the number of axes of basic taste is limited to 2 or 3 as described above, so the result of principal component analysis is also It is only necessary to look at the second principal component up to two degrees, that is, it is easy to understand if it is considered in two or three dimensions.

Next, using 12 kinds of liquor as variables and 16ch taste sensor as a sample, consider differences in characteristics between channels of the taste sensor. Figure of results of principal component analysis
Shown in 11. In the figure, the closer the channels are, the more similar the sensor is, and the farther the channel is, the more different the sensor is. It can be seen from FIG. 11 that the groups are broadly classified into three groups: a group surrounded by ellipses including 1, 2, 3 channels, etc., a group of 4, 13 channels, and a group of 6 channels. From each group, select 4ch, 6ch, and 1ch far from them. Principal component analysis is performed using these three sensors as variables and 12 kinds of sake as samples. The result is shown in FIG. Compare the case of using all 16 channels (Fig. 9) and the case of using the above three channels (Fig. 10). Obviously the latter is the second
The contribution rate of the main component is large and represents a lot of information. In the former case, the contribution ratios of the first and second principal components are 81.6%,
While 10.5% is represented by only the first main component, in the latter case, the contribution ratio of the second main component is high at 61.2% and 27.0%. The more that is, the easier it is to identify. In the former case, the characteristics of most of the 16 sensors belong to the group including 1ch, so the analysis was performed only with the characteristics of this group, so the information concentrates only on the first principal component. did. In other words, it is evaluated that the information held by a small number of channels of other groups is ignored and that there is only one-dimensional information. However, it is possible to extract hidden information by analyzing the characteristics of the channels of the taste sensor by the principal component analysis in advance and selecting the channels and principal component analysis of the data based on the difference in the characteristics. Although the output value of the sensor is directly handled here, the principal component analysis may be performed after normalization. At this time, when the measurement error is largely different for each sensor channel, the sensor output is divided by the measurement error (standard deviation of repetition), and the ratio of the output value error is kept constant. If the sensitivities of the sensors are significantly different, the sensor output is divided by the vector length to standardize the vector length to 1, that is, the sensitivities are made constant.

[0034]

According to the first aspect of the present invention, the measurement data of horse mackerel by the taste sensor using the lipid membrane having a plurality of channels is subjected to principal component analysis to position the solution to be measured in a two-dimensional or three-dimensional space. Therefore, it became easier to understand the difference in horse mackerel between the solutions to be measured. According to the second invention, a taste sensor for determining which taste sensor data is used for the principal component analysis prior to the principal component analysis for positioning the solution to be measured in the two-dimensional or three-dimensional space. Since it was decided to perform the principal component analysis using the sample solution as a variable as the sample, it became easier to understand the difference in horse mackerel between the sample solutions.

[Brief description of drawings]

FIG. 1 is a flowchart showing a horse mackerel feature extraction method of the first invention.

FIG. 2 is a flowchart showing a horse mackerel feature extraction method of the second invention.

FIG. 3 is a diagram showing an output pattern of a sensor when channels are plotted on the horizontal axis.

FIG. 4 is a schematic diagram showing a lipid membrane by an expression method used in a chemical design method.

FIG. 5 is a schematic view of a taste sensor, (a) is a front view,
(B) is a sectional view.

FIG. 6 is a diagram showing a horse mackerel measurement system.

FIG. 7 is a schematic diagram showing a monomolecular film by an expression method used in a chemical design method.

FIG. 8 is a diagram showing the output of the sensor for sake on a radar chart.

FIG. 9 is a diagram showing a principal component analysis result for sake using 1 ch to 16 ch as variables.

FIG. 10: Only 1ch, 4ch, and 6ch are used for variables,
The figure showing the result of principal component analysis for sake.

FIG. 11 is a diagram showing a difference in sensor characteristics as a result of principal component analysis when sake is used as a variable and a sensor is used as a sample.

[Explanation of symbols]

 1 Base Material (Substrate) 2 Electrode 3 Lipid Membrane 4 Buffer Layer 5 Lead Wire 7 Base Membrane 11 Solution to be Measured 12 Container 13 Taste Sensor Array 14 Each Lipid Membrane (Indicated by Black Point) 15 Reference Electrode 16 Buffer Layer 17 Lead Wire 18 Buffer amplifier 20 Analog switch 21 AD converter 22 Microcomputer 23 XY recorder 24 Ground potential 31 Lipid molecule group 31 'Lipid molecule group 32 Membrane member 33 Matrix 36 Amphiphile molecule group or bitter substance molecule group 37 Amphiphilic Molecule or molecule of bitter substance

Claims (2)

[Claims] 1. A step of measuring, for each channel, data relating to aji of a plurality of solutions to be measured using a taste sensor having a plurality of channels using a plurality of types of lipid membranes, and the plurality of solutions to be measured. Using the plurality of channels as variables and principal component analysis of the data to determine the components to be the first axis, the second axis, and the third axis, and the first axis and the second axis or the third axis. On the plane determined by
A method for extracting a horse mackerel feature, which comprises the step of positioning the solution to be measured based on the data converted in the process of principal component analysis. 2. A step of measuring data relating to aji of a plurality of solutions to be measured for each channel by using a taste sensor having a plurality of channels using a plurality of types of lipid membranes, and sampling the plurality of channels. And a plurality of solutions to be measured as a variable, a step of performing the principal component analysis of the data, a step of selecting a desired plurality of channels from the plurality of channels from the result of the principal component analysis, the plurality of solutions to be measured Using the selected plurality of channels as variables and performing principal component analysis on the data to determine the components to be the first axis, the second axis, and the third axis, and the first axis and the second axis, or On the plane determined by the third axis,
A method for extracting a horse mackerel feature, which comprises the step of positioning the solution to be measured based on the data converted in the process of principal component analysis.