JP7703182B2 - Texture estimation method - Google Patents

Description

The present invention relates to a food texture estimation method that can estimate the food texture.

Traditionally, the deliciousness of food has been said to be comprised of three elements: flavor, aroma, and texture. Of these, texture is made up of extremely delicate and complex information. Texture, which is determined by a complex of factors such as stickiness and melt-in-the-mouth feel, is particularly difficult to quantify. For this reason, this type of texture is often evaluated by sensory evaluation by multiple evaluators.

On the other hand, a food texture evaluation method such as that described in Patent Document 1 is known. In this method, first, a food product is placed on a stage below a wedge-shaped plunger of a measuring device, and the tip of the wedge-shaped plunger is pressed against the food product to measure the load data.

Then, the average values of the feature quantities of the load data obtained from the sample load data using the Texture Profile Analysis (TPA) method, namely "hardness," "brittleness," "elasticity," and "gumminess," are obtained and used as model data for creating a standard food texture model (Patent Document 1/paragraph 0054, Figures 1 and 4).

Patent application No. 2021-032656

However, it is difficult to accurately evaluate texture from load data alone, especially textures that depend on individual sensations such as "crispy," "crunchy," and "crunchy," and the data does not always match sensory evaluation.

The present invention was made in consideration of these circumstances, and aims to provide a texture estimation method that can accurately estimate the texture of various foods.

In order to achieve the above object, the present invention provides a food texture estimation method for estimating the food texture of a food to be evaluated, comprising the steps of:
the step of performing a sensory evaluation on a model food for food texture estimation and obtaining a sensory evaluation value (λ); the step of obtaining measurement data (A) including acoustic data when the model food is pressed using a pressing device that presses food; the step of performing a fast Fourier transform on the acoustic data contained in the measurement data (A) for each arbitrary time interval to obtain a power spectrum (PS _A ) for each frequency or for each frequency band; the step of obtaining a plurality of feature quantities (P _A ) other than the power spectrum from the measurement data (A) that are specific to the model food; the step of calculating a Z-score from the power spectrum (PS _A ) and the average value and standard deviation of the feature quantities (P _A ); and the step of creating a standard food texture model corresponding to a texture consistent with the sensory evaluation value (λ) using the power spectrum (PS _A ) and the feature quantities (P _A ) for which the Z-score falls within a predetermined range.

In the food texture estimation method of the present invention, first, a sensory evaluation is performed on a model food to obtain a sensory evaluation value (λ). The sensory evaluation value (λ) is used to create a standard food texture model. Next, a pressing device is used to obtain measurement data (A) of the model food. The measurement data (A) includes acoustic data when pressed, but also other data (load data, etc.).

For the acoustic data of the measurement data (A), a fast Fourier transform is performed to obtain a power spectrum (PS _A ), which becomes a feature specific to the food. A _Z -score is then calculated from the average value and standard deviation of the power spectrum (PS _A ) and the feature (P _A ), and a standard texture model corresponding to the texture of the sensory evaluation value (λ) is created using the power spectrum (PS _A ) and the feature (P _A ) for which the Z-score falls within a predetermined range. In this way, by using the standard texture model obtained by statistically processing the power spectrum (PS _A ) and the feature (P A ), the texture of various evaluation foods when eaten can be accurately estimated.

It is preferable that the food texture estimation method of the present invention comprises the steps of: acquiring measurement data (B) including acoustic data when the food for evaluation is pressed using the pressing device; performing a fast Fourier transform on the acoustic data contained in the measurement data (B) for each arbitrary time interval to acquire a power spectrum (PS _B ) for each frequency or for each frequency band; acquiring a plurality of feature quantities (P _B ) other than the power spectrum that are specific to the food for evaluation from the measurement data (B) of the food for evaluation; and evaluating the power spectrum (PS _B ) and the feature quantities (P _B ) using the standard food texture model to estimate the texture of the food for evaluation.

According to this configuration, measurement data (B) including acoustic data is acquired for the evaluation food, and a power spectrum (PS _B ) and feature quantity (P _B ) are acquired. Then, a texture evaluation of the evaluation food is performed using a standard texture model prepared in advance. This makes it possible to easily estimate the texture of the evaluation food when eaten.

In the food texture estimation method of the present invention, it is preferable that said standard food texture model is a prediction formula obtained by regression analysis using said power spectrum (PS _A ) and said feature amount (P _A ) as explanatory variables and said sensory evaluation value (λ) as a response variable.

By using the prediction formula (regression equation), the power spectrum (PS _B ) and feature quantity ( _PB ) of the evaluation food can be substituted to obtain a model evaluation value (y). This makes it possible to estimate the texture of the evaluation food when eaten using the standard texture model.

In addition, in the food texture estimation method of the present invention, it is preferable to use a Gaussian process regression model for the regression analysis.

The Gaussian process regression model is a nonlinear regression model, but there is a nonlinear relationship between the acoustic features (power spectrum) and the sensory evaluation value. For this reason, by performing regression analysis using multiple features, including acoustic data, it is possible to obtain results that are close to sensory evaluation.

In addition, in the food texture estimation method of the present invention, it is preferable that the measurement data (A) includes load data and vibration data obtained when the model food is pressed.

The measurement data (A) includes load data and vibration data when the model food is pressed by the pressing device, so that the characteristic quantity (P _A ) specific to the food can be obtained from these data, thereby improving the accuracy of the created standard food texture model.

In addition, in the food texture estimation method of the present invention, it is preferable that the measurement data (B) includes load data and vibration data obtained when the food for evaluation is pressed.

The measurement data (B) also includes load data and vibration data when the evaluation food is pressed with the pressing device. Therefore, a feature quantity (P _B ) specific to the food can be obtained from these data, and the food texture can be estimated when the evaluation food is eaten using the standard food texture model.

The present invention makes it possible to estimate the texture of food with high accuracy.

FIG. 1 is a schematic diagram illustrating a food texture estimation method according to an embodiment of the present invention. FIG. 2 is a diagram for explaining a measurement system used to measure the weight, sound, etc. of food. FIG. 2 is a diagram illustrating an internal configuration of a pressing device. 1A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 1-A), and FIG. 1B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 1-B). 2A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 2-A), and FIG. 2B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 2-B). 1A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 3-A), and FIG. 1B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 3-B). 4A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 4-A), and FIG. 4B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 4-B). 5A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 5-A), and FIG. 5B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 5-B). 6A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 6-A), and FIG. 6B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 6-B). 7A is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 7-A), and FIG. 7B is a diagram for explaining the analysis result of the fast Fourier transform of acoustic data (Sample 7-B). A diagram explaining the estimation results from regression analysis of the Gaussian process regression model (with acoustic data). A diagram explaining the estimation results from regression analysis of the Gaussian process regression model (no acoustic data). A diagram explaining the estimation results from regression analysis of the linear regression model (with acoustic data). A diagram explaining the estimation results from regression analysis of the linear regression model (no acoustic data).

Below, we will explain the food texture estimation method according to an embodiment of the present invention with reference to the drawings.

Fig. 1 shows a schematic diagram of the food texture estimation method of the present invention. First, model foods F _M (usually 6 to 7 types) are used to create a standard texture model M for evaluating evaluation foods F _E. It is advisable to select multiple foods for the model foods F _M that have a texture similar to that of the evaluation foods F _E.

First, a "sensory evaluation value λ" for the model food F _M is obtained through a sensory evaluation by an evaluator (step S01). When performing a sensory evaluation of chicken nuggets, the same chicken nuggets as the model food F _M are used. Then, the evaluator chews the model food F _M and evaluates the intensity of the texture, such as "crispy,""crunchy," and "crunchy," using a visual analogue scale (VAS) or the like.

Next, the model food F _M is measured by the pressing device 10 (step S02). The "measurement data A" obtained by pressing the model food F _M with the pressing device 10 is mainly load data, vibration data, and acoustic data of the model food F _M , and the "feature quantity P _A " described below is obtained from the load data and vibration data. Note that if the model food F _M is the same as the evaluation food F _E described below, the "measurement data A" may utilize part of the "measurement data B" obtained by pressing the evaluation food F _E.

The "feature amount P _A " obtained from the load data includes the hardness, brittleness, adhesive strength, stickiness, depression, cohesiveness, elasticity, gumminess, chewiness, and the DTW distance described below of the model food F _M.

Furthermore, as the "feature amount P _A " from the vibration data, the model food F _M 's first maximum moving average, second maximum moving average, positive pulse average, positive pulse standard deviation, positive pulse variance, positive pulse skewness, positive pulse kurtosis, number of positive pulses, number of negative pulses, average distance between pulses, maximum distance between pulses, sum of amplitudes in 1/1 octave bands (V1 to V12), and similarity to an arbitrary template waveform are obtained. In this embodiment, the similarity is quantitatively evaluated using dynamic time warping (DTW). The dissimilarity or similarity that can be obtained by calculating the DTW distance with the waveform to be compared can be used as one of the feature amounts of the food.

The DTW distance is the cumulative value of the distances found by comparing the distances between the elements of two waveforms in a brute-force manner and then finding the path that has the shortest distance between those points. The more similar the two pieces of data are, the smaller the DTW distance (the higher the similarity) and the closer the waveform shapes are.

This time, for all "measurement data A", average data of model foods F _M was created as template waveforms using the DBA (DTW Barycenter Averaging) method. By using the average data, the difference from the actual texture can be quantified as the DTW distance. Furthermore, in this embodiment, seven types of model foods F _M are used. Therefore, the similarity to a total of 14 template waveforms of 7 loads and 7 vibrations is calculated, and the total of 14 similarities are calculated as feature quantities.

In addition, the features obtained from the acoustic data are the maximum volume of the model food F _M , the maximum value of the first moving average, the maximum value of the second moving average, and the average of the A-weighting (S1 to S21) in a specific frequency band (1/3 octave band).

The "A-weighting" is a feature of sound weighted by frequency to match the human hearing. To obtain the A-weighting, a fast Fourier transform (FFT) is executed by a computer 40 (described later) (step S03). This analysis converts the "measurement data A" of the time series data into a frequency axis, and the "power spectrum PS _A " (sound feature) for each frequency can be obtained.

It is also preferable to limit the measurement time range using a pseudo rectangular window with a width of 1 second and obtain the power spectrum (average power in 1/3 octave band). This is because, depending on the type of food, sound is generated only for a short period of time from when the food comes into contact with the probe or when the food is maximum compressed, and sound due to breakage or compression is not recorded for most of the measurement period, preventing feature extraction when averaging sound data over the measurement time range.

Then, the "power spectrum PS _A " and the "feature quantity _PA " (load and vibration) are used to calculate the Z-score (step S04). The Z-score is a method for normalizing each feature quantity, and in order to calculate the Z-score, the average value and standard deviation are calculated for each feature quantity. The Z-score (z) is given by the following formula (1), where the average value is x _AVE and the standard deviation is s.
z=(x- _xAVE )/s...(1)

Then, variables whose Z scores fall within a predetermined range, i.e., variables whose Z scores fall outside the predetermined range are excluded from the "power spectrum PS _A " and "feature quantity _PA ", and a "standard food texture model M" is created using the remaining variables (step S05). The "standard food texture model M" is a prediction equation (regression equation) for evaluating the food texture of the evaluation food F _E. In this case, the variables used in the regression may be the selected power spectrum PS _A and feature quantity _PA , which are used as they are without being converted, or the Z scores of these variables may be used.

When performing linear regression analysis, the standard food texture model M is given by the following formula (2) where the explanatory variable x is the power spectrum PS _A and the feature amount _PA (x ₁ , x ₂ , ..., x _k ) and the objective variable y is the sensory evaluation value λ (estimated value).
y=β ₀ +β ₁ x ₁ +β ₂ x ₂ +...+β _k x _k ...(2)
Here, β ₀ is a constant term, and β ₁ to β _k are coefficients of x ₁ to x _k , respectively.

Specifically, the power spectrum PS _A , feature amount _PA and sensory evaluation value λ are substituted into the above formula (2) using a program or the like, and the coefficients β ₀ , β ₁ , β ₂ , ..., β _k are obtained by regression analysis to obtain the standard food texture model M. In this example, the standard food texture model M was created using the power spectrum PS _A and feature amount _PA whose Z scores fall within a predetermined range, but instead of the Z scores, methods such as deviation values (average: 50, standard deviation: 10), deviation IQ (average: 100, standard deviation: 15 or 16) may also be used.

Next, the evaluation food F _E is measured by the pressing device 10 (step S06). Seven types of chicken nuggets with different batter textures were used as the evaluation food F _E. This allows for the "measurement data B" of the evaluation food _F E to be obtained. The "measurement data B" is composed of the load data, vibration data, and sound data of the evaluation food F _E.

From the "measurement data B", a "feature value P _B " relating to the load data and vibration data is obtained. In addition, as feature values from the acoustic data, the maximum volume, the first maximum value of the moving average, and the second maximum value of the moving average of the evaluation food F _E are obtained. Furthermore, a fast Fourier transform (FFT) is performed to obtain the A-weighted "power spectrum (PS _B )" in a specific frequency band (1/3 octave band) (step S07).

Using the "power spectrum PS _B " and "feature quantity P _B " (load and vibration), the texture of the evaluation food F _E is estimated by a previously created standard food texture model M (step S08). This results in a "model evaluation value y." Note that, if the standard food texture model M is obtained using the Z-scores of each variable as variables, the power spectrum PS _B and feature quantity P _B are also converted into Z-scores.

Next, referring to FIG. 2, a measurement system 100 used to measure the load of food, etc., will be described. The measurement system 100 is composed of a pressing device 10 that presses the food, an integrated board 20 on which various sensors are attached, a microphone 30 that captures the sound generated when the food is pressed, and a computer 40 that analyzes the sound data from the microphone 30.

The pressing device 10 shown in Figure 2 is a device having a structure equivalent to human teeth, periodontal ligament, etc., and uses various sensors to measure the information perceived by a human when chewing the evaluation food F _E in order to evaluate the texture of the evaluation food F _E.

The measurement conditions this time were a pressing speed of 10 (mm/s) and a compression rate of 90% for the food. This operation was repeated 20 times for each sample of the evaluation food F _E (seven types of chicken nuggets).

The pressing device 10 is composed of a probe 11 that corresponds to a tooth, a spring 12 that reproduces the elasticity of the periodontal ligament, a linear slider 13, a cylindrical shaft 14, a base plate 18, and an integrated substrate 20. The probe 11, spring 12, and linear slider 13 are housed in the cylindrical shaft 14, supported below by the base plate 18, and placed on the integrated substrate 20.

The probe 11 is cylindrical and made of a hard material, with its tip protruding from a cylindrical shaft 14, and moves up and down when subjected to a force Fz in the Z-axis direction. A permanent magnet 15 is embedded inside the probe 11, and a magnetic resistance element 21 and an inductor element 22 provided on an integrated substrate 20 detect the change in magnetism when the food product F _E to be evaluated is pressed.

Instead of the spring 12, the pressing device 10 may also have an intermediate layer made of an elastic elastomer (natural rubber, synthetic rubber, etc.). The linear slider 13 is provided to reduce friction between the probe 11 and the cylindrical shaft 14. The permanent magnet 15 may be a neodymium magnet (e.g., 15 mm in diameter, 1.5 mm in thickness, and 115 mT in surface magnetic flux density).

The magnetic resistance element 21 can detect the change in position (amount of displacement) of the permanent magnet 15. The change in position when the evaluation food F _E presses the probe 11 is converted by a calibration formula and recorded as "load data."

The inductor element 22 can detect the induced electromotive force caused by the dynamic change in magnetic flux density that occurs when the permanent magnet 15 is displaced. The magnitude of the induced electromotive force depends on the speed of change in magnetic flux density rather than the amount of change. The change in magnetic flux density when the evaluation food F _E presses the probe 11 is recorded as the "vibration data" of the probe 11.

Two magnetoresistance elements 21 are provided on the integrated substrate 20, sandwiching the inductor element 22. The permanent magnet 15 is located below the probe 11 (on the spring 12 side) and at the center of the circle, so when the food under evaluation presses against the probe 11, the position of the permanent magnet 15 changes. Therefore, by converting the output voltage of the magnetoresistance element 21, which measures the magnetic field at that time, using a calibration formula, it is possible to detect the load that the probe 11 receives from the food under evaluation.

One inductor element 22 is provided on the integrated substrate 20 at a position corresponding to the probe 11. This makes it possible to detect changes (vibrations) in the magnetic flux density of the permanent magnet 15 located almost directly above the inductor element 22. The number of magnetic resistance elements 21 and inductor elements 22 can be changed as appropriate. The signals detected by the magnetic resistance elements 21 and inductor elements 22 are converted by an AD converter 23 (not shown) and transmitted to the control unit.

In addition, the sound produced when the evaluation food F _E presses the probe 11 is picked up by the microphone 30 and recorded as "acoustic data". The microphone 30 only needs to have the performance to recognize frequencies audible by humans. The "acoustic data" picked up by the microphone 30 is transmitted to the computer 40, where the above-mentioned fast Fourier transform (FFT) is performed.

Next, the internal configuration of the pressing device 10 of this embodiment will be described with reference to FIG.

The pressing device 10 is equipped with a texture sensor consisting of a probe 11, a spring 12, a linear slider 13, a cylindrical shaft 14, a permanent magnet 15, a magnetic resistance element 21, an inductor element 22 and an AD converter 23, a motor driver (manufactured by SMC) 16 and an electric actuator (manufactured by SMC) 17.

The texture sensor is attached to the end of the electric actuator 17. The electric actuator 17 is driven by a control signal from the computer 40 via the motor driver 16 of the pressing device 10.

In the actual measurement system 100, the probe 11 of the pressing device 10 is positioned so that it faces vertically downward, and the evaluation food F _E is pressed by the probe 11. Note that the measurement range is set to the output load when the electric actuator 17 presses the probe 11 and compresses the food by 90%.

At this time, the food texture sensor outputs an output voltage of the magnetoresistance element 21 (magnetoresistance voltage: voltage signal A) and an output voltage of the inductor element 22 (inductor voltage: voltage signal B). Voltage signal A and voltage signal B are sent to the AD converter 23 and converted into measurement values. Furthermore, the measurement values are sent to the computer 40. The sound picked up by the microphone 30 is also sent to the computer 40.

Next, the analysis results of the fast Fourier transform of the acoustic data will be explained with reference to Figures 4 to 10.

FIG. 4(a) shows the analysis results of "Sample 1-A (first 10 compressions)" of evaluation food F _E , and FIG. 4(b) shows the analysis results of "Sample 1-B (last 10 compressions)" of evaluation food F _E. The vertical axis of the graph indicates frequency (kHz), and the horizontal axis indicates step time. There are 11 step times from -1 to 10 seconds, with the start of the first compression being approximately 0 seconds (Time step "2") and the start of the second compression being approximately 5 seconds (Time step "7"). The power spectrum intensity of each step time and frequency band is visualized on the color scale on the right, with the greater the power spectrum intensity, the brighter the color.

Figures 5, 6, 7, 8, 9, and 10 show the analysis results for Sample 2, Sample 3, Sample 4, Sample 5, Sample 6, and Sample 7, respectively, and, like Figure 4, are shown divided into the first 10 compressions and the last 10 compressions.

For example, as shown in "Data S44" in Figure 6(a), a strong power spectrum with a deep white color was obtained near time steps "2 to 4" and near time steps "7 to 9." Also, if the sample is different, the tendency of the obtained frequency (height, width, etc.) will also be different.

Sounds are not always generated during chewing, and the timing at which characteristic sounds are generated varies for each food. For this reason, it is important to accurately capture the sound at the appropriate timing. In this embodiment, the average power spectrum of 1/3 octave of the acoustic data immediately after the start of chewing (directly above time step "2"), where the differences between the samples were most apparent, was used. In "Sample 3" (see Figure 6), "Sample 4" (see Figure 7), and "Sample 7" (see Figure 10), high power spectra were obtained over a relatively wide range.

Next, we will explain the results of the regression analysis with reference to Figures 11 to 14.

Here, among the features other than the DTW distance, the gumminess and chewiness of the "load data", the positive pulse standard deviation, positive pulse variance, positive pulse skewness, number of positive pulses, and part of the sum of the amplitude in the 1/1 octave band of the "vibration data" (V2 to V12), and the maximum volume and part of the average A-weighting in the 1/3 octave band of the "sound data" (S4, S6 to S8, S10, S14, S16 to S18, S20) were removed by calculating the variance inflation factor (VIF) in order to reduce multicollinearity in the regression analysis.

For each remaining feature, the average and standard deviation were calculated for each sample, and the Z-score of each feature was calculated. In addition, the sum of the Z-scores (absolute values) of each feature, excluding the DTW distance, for each sample was calculated, and the eight data with the largest values were excluded as outliers, and the remaining 12 data were used.

First, Figures 11 and 12 show the results of regression analysis using a Gaussian process regression model (GP). Table 1 also shows the RMSE (Root Mean Squared Error) value when regression analysis was performed using the Gaussian process regression model. The Gaussian process regression model is a nonlinear regression model (expressing a variable as a quadratic or cubic expression of other variables), and is effective when fitting is not possible using a linear regression model.

The vertical axis of the graph shows the texture evaluation value, and the horizontal axis shows the sample index number. "O" indicates the estimated value from the regression analysis, and "+" indicates the target value, which is equal to the sensory evaluation value. The pair of upper and lower waveforms indicates the confidence interval, and the estimated value that falls between the two waveforms can be said to be a statistically reliable value.

Figure 11 shows the results of regression analysis using each feature, including "acoustic data," and shows the textures of "crispy," "crunchy," and "crunchy" from top to bottom. As shown in Table 1, the RMSE values were 1.40 for "crispy," 0.97 for "crunchy," and 1.45 for "crunchy." The RMSE value is the root mean square error for each texture, and the lower the value, the higher the estimation accuracy.

Figure 12 shows the results of regression analysis using each feature excluding "acoustic data", and a comparative analysis was conducted to investigate the influence of acoustics. As shown in Table 1, the RMSE values were 1.48 for "crispy", 1.01 for "crunchy", and 1.49 for "crunchy". As all of these values were higher than the results including "acoustic data", it was concluded that regression analysis using each feature including "acoustic data" allows for more accurate texture estimation.

Next, Figures 13 and 14 show the results of regression analysis using a linear regression model (LM). Table 2 also shows the RMSE value when regression analysis was performed using the linear regression model. A linear regression model is a regression model that analyzes a variable by expressing it as a linear equation of other variables.

The vertical axis of the graph shows the evaluation value, and the horizontal axis shows the sample index number. Also, "◯" indicates the estimated value from the regression analysis, and "+" indicates the target value, which is equal to the sensory evaluation value.

Figure 13 shows the results of regression analysis using each feature, including "acoustic data," and shows the textures of "crispy," "crunchy," and "crunchy" from top to bottom. The RMSE values were 1.90 for "crispy," 1.29 for "crunchy," and 1.51 for "crunchy."

Figure 14 shows the results of regression analysis using each feature excluding "acoustic data", and a comparative analysis was performed to examine the influence of acoustics. The RMSE values were 1.90 for "crispy", 1.88 for "crunchy", and 1.52 for "crunchy". For "crunchy", a lower value was obtained when using features including "acoustic data", which suggests that the accuracy of estimating texture has improved.

From the RMSE values shown in Tables 1 and 2, it was found that the texture of a food can be estimated with high accuracy by obtaining the acoustic features (power spectrum) in addition to the load and vibration features of the food, and performing regression analysis using a Gaussian process regression model.

Regression analysis may be performed using a nonlinear model other than the Gaussian process regression model. In addition, the method for excluding some feature values as outliers may be determined appropriately based on the Z-score value, the total number of valid values, etc. Of course, regression analysis may be performed without outliers.

10...Pressing device, 11...Probe, 12...Spring, 13...Linear slider, 14...Cylindrical shaft, 15...Permanent magnet, 16...Motor driver, 17...Electric actuator, 18...Base plate, 20...Integrated board, 21...Magnetic resistance element, 22...Inductor element, 23...A/D converter, 30...Microphone, 40...Computer, 100...Measurement system, M...Standard food texture model.

Claims (2)

A food texture estimation method for estimating the food texture of an evaluation food, comprising:
conducting a sensory evaluation on the model food for food texture estimation and obtaining a sensory evaluation value (λ);
A step of acquiring measurement data (A) including acoustic data , vibration data, and load data when the model food is pressed using a pressing device that presses food;
performing a fast Fourier transform on the acoustic data included in the measurement data (A) for each arbitrary time interval to obtain a power spectrum (PS _A ) for each frequency or for each frequency band;
acquiring the vibration data and the load data from the measurement data (A) as a feature quantity (P _A );
calculating a Z-score from the power spectrum (PS _A ) and the average value and standard deviation of the feature quantity (P _A );
creating a standard food texture model according to the food texture that is consistent with the sensory evaluation value (λ) using the power spectrum (PS _A ) and the feature amount (P _A ) for which the Z-score falls within a predetermined range;
Equipped with
the food texture standard model is a prediction formula obtained by regression analysis using the power spectrum (PS _A ) and the feature amount (P _A ) as explanatory variables and the sensory evaluation value (λ) as a response variable,
13. A food texture estimation method comprising : a Gaussian process regression model being used for the regression analysis . a step of acquiring measurement data (B) including acoustic data , vibration data, and load data when the food product for evaluation is pressed using the pressing device;
performing a fast Fourier transform on the acoustic data contained in the measurement data (B) for each arbitrary time interval to obtain a power spectrum (PS _B ) for each frequency or for each frequency band;
acquiring the vibration data and the load data from the measurement data (B) of the food product under evaluation as a feature quantity (P _B );
a step of evaluating the power spectrum (PS _B ) and the feature amount (P _B ) using the standard food texture model to estimate the food texture of the evaluation food;
The food texture estimating method according to claim 1, further comprising: