Steaming degree estimation device for tea leaves, steaming degree estimation method, and tea leaf manufacturing method

Abstract

To provide an apparatus and a method for estimating the degree of steaming of tea leaves, capable of objectively estimating the degree of steaming of steamed leaves, and to provide a method for producing tea leaves.SOLUTION: A steaming degree estimation device for tea leaves includes first measurement means for measuring light intensities of reflected light at a plurality of specific wavelengths by irradiating a plurality of tea leaf samples each having a known steaming degree with light in a visible region, prediction model creation means for creating a steaming degree prediction model representing a correspondence relationship between the light intensities of the reflected light at the plurality of specific wavelengths measured by the first measurement means and the known steaming degrees in the plurality of tea leaf samples, storage means for storing the steaming degree prediction model created by the prediction model creation means, and second measurement means for measuring a light intensity of reflected light at a specific wavelength by irradiating target tea leaves whose steaming degree is to be estimated with light in a visible region. And a steaming degree estimation means for obtaining the steaming degree of the target tea leaves corresponding to the light intensity of the reflected light measured by the second measurement means.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a steaming degree estimation device for estimating the degree of steaming of steamed leaves, a steaming degree estimation method, and a tea leaf manufacturing method. [Background technology]

[0002] Immediately after harvesting, tea leaves begin to deteriorate due to the action of oxidizing enzymes. For this reason, tea production involves a steaming process in which harvested tea leaves are steamed to halt the deterioration process. It is known that the taste and aroma of the product change depending on the steaming level, which indicates the degree of steaming during this steaming process. For example, lightly steamed tea leaves with a low steaming level have a bitter, greenish, and fresh aroma, whereas deeply steamed tea leaves with a high steaming level have a sweet, mellow taste but a weaker aroma. Therefore, steaming level is an important factor affecting product quality (Non-Patent Document 1). Furthermore, tea production involves multiple drying processes to dry the steamed leaves that have retained moisture after the steaming process. It is preferable to change the conditions of these drying processes (such as the temperature of the hot air, the force with which the steamed leaves are kneaded, the air volume, and the working time) depending on the steaming level. For this reason, it is desirable to accurately estimate the steaming level of steamed leaves.

[0003] Traditionally, steaming degree has been estimated by experienced technicians through their empirical and sensory judgment (sensory evaluation) based on the aroma, color, shape, and texture (touch) of the steamed leaves. However, sensory evaluation is influenced by the environment of the test site and the person's physical and mental condition. Furthermore, there are very few experienced technicians who can accurately estimate the steaming degree of tea leaves. Furthermore, training such skilled technicians is not easy, and the transfer of skills has been extremely difficult. For these reasons, there has been a demand for the establishment of a scientific method for objectively estimating steaming degree.

[0004] A common scientific method for estimating steaming degree is to estimate it based on the steaming time in the steaming process. This method is based on the idea that the steaming temperature in the steaming process is constant at 100°C, and that the steaming degree can be calculated as the product of temperature and steaming time (Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Yuichi Shibata, "Theory and Practice of Machine-Made Tea," Process Management in Accordance with Tea Leaves and the Environment, Agriculture, Forestry and Fishing Villages Cultural Association, First Edition, September 30, 2006 Summary of the Invention [Problem to be solved by the invention]

[0006] However, in recent years, attempts have been made to improve the quality of tea leaves in the steaming process by steaming them at lower temperatures or by changing the stirring conditions of the tea leaves. Therefore, evaluating the steaming degree based solely on the steaming time in the steaming process is no longer in line with the actual situation.

[0007] The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a tea leaf steamness degree estimation device, a steamness degree estimation method, and a tea leaf manufacturing method that can objectively and accurately estimate the steamness of steamed leaves. [Means for solving the problem]

[0008] The tea leaf steamness estimation device of the present invention comprises a first measurement means for irradiating a tea leaf sample having a known steamness with light in the visible and near-infrared regions and measuring the light intensity of the reflected light at a plurality of wavelengths; a prediction model creation means for creating a steamness prediction model that shows the correspondence between the light intensity of the reflected light at the plurality of wavelengths measured by the first measurement means and the known steamness of the tea leaf sample; a storage means for storing the steamness prediction model created by the prediction model creation means; a second measurement means for irradiating target tea leaves whose steamness is to be estimated with light in the visible and near-infrared regions and measuring the light intensity of the reflected light at a plurality of wavelengths; and a steamness estimation means for determining the steamness of the target tea leaves that corresponds to the light intensity of the reflected light measured by the second measurement means, from the steamness prediction model stored in the storage means.

[0009] There is a certain degree of correlation between the light intensity of reflected light at multiple wavelengths when light in the visible and near-infrared regions is irradiated and the steaming level. By using a model created by statistical learning or calculation using the reflection intensities of reflected light at these multiple wavelengths as explanatory variables and the steaming level as the target variable, it is possible to estimate the steaming level of the target tea leaves from the reflection intensity of the reflected light from the target tea leaves.

[0010] The steamness estimation device for tea leaves according to the present invention also comprises a first measurement means for irradiating a plurality of tea leaf samples, each having a known steamness level, with light in the visible range and measuring the light intensities of reflected light at a plurality of specific wavelengths; a prediction model creation means for creating a steamness prediction model that shows the correspondence between the light intensities of reflected light at the plurality of specific wavelengths measured by the first measurement means and the known steamness levels of the plurality of tea leaf samples; a storage means for storing the steamness prediction model created by the prediction model creation means; a second measurement means for irradiating target tea leaves, the steamness of which is to be estimated, with light in the visible range and measuring the light intensities of reflected light at the specific wavelengths; and a steamness estimation means for determining, from the steamness prediction model stored in the storage means, the steamness of the target tea leaves that corresponds to the light intensities of reflected light measured by the second measurement means.

[0011] Because tea leaves change color when steamed, it is possible to use the color of steamed tea leaves as a predictor of steaming level. However, while there is some correlation between tea leaf color and steaming level, the degree of correlation is not very high. The inventors of the present application discovered that there is a correlation between steaming level and the reflection intensity at wavelengths (specific wavelengths) that exhibit characteristic reflection intensities among the light reflected from tea leaves when irradiated with light in the visible range. Therefore, by using a steaming level prediction model created from the correspondence between the reflection intensity of reflected light at specific wavelengths and steaming level, it is possible to reliably estimate the steaming level of tea leaves. There is a high correlation between the light intensity of reflected light at multiple specific wavelengths when irradiated with light in the visible range and steaming level. By using a model created by machine learning or calculation using the reflection intensity of reflected light at these specific wavelengths as a feature variable and the steaming level as a target variable, it is possible to reliably estimate the steaming level of target tea leaves from the reflection intensity of the reflected light from the target tea leaves.

[0012] It is preferable that the predictive model creation means is a means for creating a machine learning model created by deep learning, in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured by the first measurement means are used as explanatory variables and the known steaming degrees of the multiple tea leaf samples are used as objective variables. By using such a machine learning model created by deep learning as a predictive model, the steaming degree of steamed leaves can be estimated with high accuracy.

[0013] It is also preferable that the prediction model creation means is means for creating a multiple regression model having a multiple regression prediction formula in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured by the first measurement means are used as explanatory variables and the known steaming degree of the tea leaf sample is used as the objective variable. By using a multiple regression model created by such a multiple regression prediction formula as a prediction model, highly reliable estimation can be performed, and the steaming degree can be estimated in a short time, which is expected to reduce costs.

[0014] In this case, it is also preferable that the prediction model creation means includes a narrowing-down means for selecting a small number of wavelengths from multiple specific wavelengths to narrow down the effective explanatory variables. By narrowing down the number of terms by selecting only effective explanatory variables, the multiple regression prediction formula can be significantly simplified. Furthermore, compared to models without variable selection (models with many explanatory variables and many terms in the formula), the prediction accuracy is comparable, with no significant difference. Furthermore, because the reflected light intensity at a small number of wavelengths is used, there is no need to measure the reflected light intensity at all wavelengths using an expensive multi-wavelength spectroscopic camera; instead, the reflected light intensity can be measured at low cost using an RGB camera or smartphone camera.

[0015] It is also preferable that the prediction model creation means is means for creating a partial least squares regression model having a partial least squares regression prediction formula in which the light intensities of reflected light at the multiple wavelengths measured by the first measurement means are used as explanatory variables and the known steaming degrees of the multiple tea leaf samples are used as response variables. By using a partial least squares regression model created using such a partial least squares regression prediction formula as a prediction model, highly reliable estimation can be performed.

[0016] In this case, it is also preferable that the prediction model creation means includes a division means for obtaining explanatory variables by dividing the light intensity of reflected light at multiple wavelengths measured by the first measurement means by the wavelength band average value. Since the difference in reflected light intensity caused by tea leaves harvested at different times is accounted for by the flat, stable wavelength band average value, which has little fluctuation, as the explanatory variable, prediction accuracy can be significantly improved.

[0017] It is also preferable that the specific wavelengths include wavelengths between 528nm and 550nm. Chlorophyll contained in tea leaves is sensitive to heat, and when steamed, the magnesium contained in the chlorophyll is released and it turns into pheophytin. Compared to chlorophyll, pheophytin absorbs light with wavelengths between 528nm and 550nm, so by using light with wavelengths in this range for learning, the degree of steaming of the target tea leaves can be estimated with high reliability.

[0018] It is also preferable that the specific wavelengths include wavelengths of 580 nm to 660 nm. Chlorophyll absorbs light with wavelengths of 580 nm to 660 nm more than pheophytin, so by using light with wavelengths in this range for learning, the steaming degree of the target tea leaves can be estimated with high reliability.

[0019] It is also preferable that the specific wavelengths include wavelengths of 670 nm to 730 nm. Because chlorophyll absorbs light with wavelengths of 670 nm to 730 nm more than pheophytin, by using light with wavelengths in this range for learning, the steaming degree of the target tea leaves can be estimated with high reliability.

[0020] It is also preferable that the first and second measuring means are light intensity measuring means equipped with a multi-wavelength spectroscopic camera. Unlike a general RGB camera, a multi-wavelength spectroscopic camera can measure the reflection intensity of reflected light at multiple wavelengths with high accuracy. Such a multi-wavelength spectroscopic camera can output the reflection intensity as a reflectance or an absorptance, and therefore can be suitably used as the first and second measuring means of the present invention.

[0021] It is also preferable that the second measuring means is configured to measure the light intensity of reflected light at a specific wavelength from the target tea leaves after steaming treatment in a steamer. The second measuring means, which measures the light intensity of reflected light at multiple wavelengths from the tea leaves, is provided at a position where it can measure the reflected light intensity of the tea leaves after they have passed through the steamer of the manufacturing equipment, such as the outlet position of the steamer or the conveying position between the steamer and the subsequent process. If a cooling treatment section is included in the process subsequent to the steamer, the target tea leaves may be those that have been cooled. By using tea leaves with reduced surface moisture after cooling as the target tea leaves, measurement accuracy can be further improved.

[0022] It is also preferable that the steaming degree estimation means is configured to transmit the steaming degree determined by the steaming degree estimation means to a tea leaf steamer. By transmitting the steaming degree determined by the steaming degree estimation device of the present invention to the tea leaf steamer, the steamer can be controlled reliably and appropriately.

[0023] Furthermore, according to the present invention, a method for estimating the steamness of tea leaves comprises: a first measurement step of irradiating a tea leaf sample having a known steamness level with light in the visible and near-infrared regions and measuring the light intensities of reflected light at a plurality of wavelengths; a prediction model creation step of creating a steamness prediction model that shows the correspondence between the light intensities of reflected light at the plurality of wavelengths measured in the first measurement step and the known steamness levels of the plurality of tea leaf samples; a storage step of storing the steamness prediction model created in the prediction model creation step; a second measurement step of irradiating target tea leaves whose steamness is to be estimated with light in the visible and near-infrared regions and measuring the light intensities of reflected light at a plurality of wavelengths; and a steamness estimation step of determining the steamness of the target tea leaves that corresponds to the light intensities of reflected light measured in the second measurement step, from the steamness prediction model stored in the storage step.

[0024] There is a certain degree of correlation between the light intensity of reflected light at multiple wavelengths when light in the visible and near-infrared regions is irradiated and the steaming level. By using a model created by statistical learning or calculation using the reflection intensities of reflected light at these multiple wavelengths as explanatory variables and the steaming level as the target variable, it is possible to estimate the steaming level of the target tea leaves from the reflection intensity of the reflected light from the target tea leaves.

[0025] Furthermore, according to the present invention, the method includes a first measurement step of irradiating a plurality of tea leaf samples, each having a known steaming level, with light in the visible range and measuring the light intensities of reflected light at a plurality of specific wavelengths; a prediction model creation step of creating a steaming level prediction model that shows the correspondence between the light intensities of reflected light at the plurality of specific wavelengths measured in the first measurement step and the known steaming levels of the plurality of tea leaf samples; a storage step of storing the steaming level prediction model created in the prediction model creation step; a second measurement step of irradiating target tea leaves, the steaming level of which is to be estimated, with light in the visible range and measuring the light intensities of reflected light at the specific wavelengths; and a steaming level estimation step of determining, from the steaming level prediction model stored in the storage step, the steaming level of the target tea leaves that corresponds to the light intensities of reflected light measured in the second measurement step.

[0026] Because tea leaves change color when steamed, it is possible to use the color of steamed tea leaves as a predictor of steaming level. However, while there is some correlation between tea leaf color and steaming level, the degree of correlation is not very high. The inventors of the present application discovered that, since there is a correlation between steaming level and the reflection intensity of specific wavelengths of light reflected from tea leaves when irradiated with visible light, it is possible to reliably estimate the steaming level of tea leaves by using a steaming level prediction model created from the correspondence between steaming level and the reflection intensity of light reflected at specific wavelengths. There is a high correlation between steaming level and the light intensity of reflected light at multiple specific wavelengths when irradiated with visible light. By using a model created by machine learning or calculation using the reflection intensity of reflected light at these specific wavelengths as feature variables and steaming level as the objective variable, it is possible to reliably estimate the steaming level of target tea leaves from the reflection intensity of reflected light from the target tea leaves.

[0027] It is preferable that the prediction model creation step is a step of creating a machine learning model created by deep learning, in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured in the first measurement step are used as explanatory variables and the known steaming degrees of the multiple tea leaf samples are used as objective variables. By using such a machine learning model created by deep learning as a prediction model, the steaming degree of steamed leaves can be estimated with high accuracy.

[0028] It is also preferable that the prediction model creation step is a step of creating a multiple regression model equipped with a multiple regression prediction formula in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured in the first measurement step are used as explanatory variables and the known steaming degree of the tea leaf sample is used as a response variable. By using a multiple regression model created using such a multiple regression prediction formula as a prediction model, highly reliable estimation can be performed, and the steaming degree can be estimated in a short time, which is expected to reduce costs.

[0029] In this case, it is also preferable that the prediction model creation process includes a narrowing-down process in which a small number of wavelengths are selected from multiple specific wavelengths to narrow down the effective explanatory variables. By narrowing down the number of terms by selecting only effective explanatory variables, the multiple regression prediction formula can be significantly simplified. Furthermore, compared to models without variable selection (models with many explanatory variables and many terms in the formula), the prediction accuracy is comparable, with no significant difference. Furthermore, because the reflected light intensity at a small number of wavelengths is used, there is no need to measure the reflected light intensity at all wavelengths using an expensive multi-wavelength spectroscopic camera; instead, the reflected light intensity can be measured at low cost using an RGB camera or smartphone camera.

[0030] It is also preferable that the prediction model creation step is a step of creating a partial least squares regression model having a partial least squares regression prediction formula in which the light intensities of reflected light at the multiple wavelengths measured in the first measurement step are used as explanatory variables and the known steaming degree of the tea leaf sample is used as a response variable. By using the partial least squares regression model created using such a partial least squares regression prediction formula as a prediction model, highly reliable estimation can be performed.

[0031] In this case, it is also preferable that the prediction model creation step includes a division step of obtaining explanatory variables by dividing the light intensity of reflected light at multiple wavelengths measured in the first measurement step by the wavelength band average value. Since the difference in reflected light intensity caused by tea leaves harvested at different times is accounted for by the flat, stable wavelength band average value, which has little fluctuation, as the explanatory variable, prediction accuracy can be significantly improved.

[0032] It is also preferable that the second measurement step measures the light intensity of reflected light at a specific wavelength from the target tea leaves after steaming treatment in a steamer. The means for performing the second measurement step, which measures the light intensity of reflected light at multiple wavelengths from the tea leaves, is provided at a position where the reflected light intensity of the tea leaves after passing through the steamer of the manufacturing equipment can be measured, such as the steamer outlet position or the conveying section position between the steamer and the subsequent process. If a cooling treatment section is included in the process subsequent to the steamer, the target tea leaves may be tea leaves after the cooling treatment. Measurement accuracy can be further improved by using tea leaves with reduced surface moisture after cooling as the target tea leaves.

[0033] It is also preferable to transmit the steaming degree determined in the steaming degree estimation step to a tea leaf steamer. By transmitting the steaming degree determined by the steaming degree estimation method of the present invention to the tea leaf steamer, the steamer can be controlled reliably and appropriately.

[0034] The present invention further provides a method for producing tea leaves in which the steaming level of the tea leaves is estimated by the above-mentioned method for estimating the steaming level of tea leaves. By sending the steaming level determined by the method for estimating the steaming level to a steamer used in producing the tea leaves, the steamer can be reliably and appropriately controlled, and ultimately, good quality tea leaves can be produced. [Effects of the Invention]

[0035] According to the steamness estimation device and steamness estimation method of the present invention, there is a certain degree of correlation between the light intensity of reflected light at multiple wavelengths when light in the visible and near-infrared regions is irradiated and the steamness. By using a model created by performing statistical learning or calculations using the reflection intensities of reflected light at these multiple wavelengths as explanatory variables and the steamness as the target variable, it is possible to estimate the steamness of target tea leaves from the reflection intensity of reflected light from the target tea leaves.

[0036] Furthermore, since there is a high correlation between the light intensity of reflected light at multiple specific wavelengths when light in the visible range is irradiated and the steaming level, a model generated by machine learning using the reflection intensity of reflected light at these specific wavelengths as a feature variable and the steaming level as a target variable can reliably estimate the steaming level of a target leaf from the reflection intensity of the light reflected from the target leaf. [Brief explanation of the drawings]

[0037] [Figure 1] 1 is a graph showing the spectrum of reflected light from raw tea leaves and tea leaves after steaming. [Figure 2] 1 is a graph showing the spectrum of reflected light from tea leaves with steaming time as a parameter. [Figure 3] 1 is a graph plotting correlation coefficients obtained over the entire wavelength range to show the correlation between the reflection intensity of tea leaves and steaming time. [Figure 4] 1 is a block diagram showing the overall configuration of a tea leaf steamness degree estimation device according to a first embodiment of the present invention. [Figure 5] FIG. 5 is a block diagram showing an example of the configuration of a prediction model in the steamness degree estimation device of FIG. 4. [Figure 6] 5 is a flowchart illustrating the operation of the steaming degree estimation device of FIG. 4. [Figure 7] 5 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 4 and the actually measured value. [Figure 8] 5 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 4 and the actually measured value. [Figure 9]5 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 4 and the actually measured value. [Figure 10] FIG. 10 is a block diagram showing an example of the configuration of a prediction model creation means in a tea leaf steaming degree estimation device according to a second embodiment of the present invention. [Figure 11] FIG. 11 is a block diagram showing an example of the configuration of a prediction model in the steamness degree estimation device of FIG. 10. [Figure 12] 11 is a flowchart illustrating the operation of the steamness degree estimation device of FIG. [Figure 13] 11 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 10 and the actually measured value. [Figure 14] 11 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 10 and the actually measured value. [Figure 15] 11 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 10 and the actually measured value. [Figure 16] 11 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimation device of FIG. 10 and the actually measured value. [Figure 17] FIG. 10 is a block diagram showing an example of the configuration of a prediction model creation means in a tea leaf steaming degree estimation device according to a third embodiment of the present invention. [Figure 18] FIG. 18 is a block diagram showing an example of the configuration of a prediction model in the steamness degree estimation device of FIG. 17. [Figure 19] 18 is a flowchart illustrating the operation of the steaming degree estimation device of FIG. 17. [Figure 20] 18 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimating device of FIG. 17 and the actually measured value. [Figure 21] 18 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimating device of FIG. 17 and the actually measured value. [Figure 22] 18 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimating device of FIG. 17 and the actually measured value. [Figure 23]18 is a graph showing the relationship between the estimated value of steaming time estimated by the steaming degree estimating device of FIG. 17 and the actually measured value. DETAILED DESCRIPTION OF THE INVENTION

[0038] Before describing the embodiments of the present invention, the concept of the tea leaf steaming degree estimating device of the present invention will be described.

[0039] There is no apparent difference in the reflected light spectra of raw tea leaves (first-grade tea) and tea leaves (first-grade tea) steamed for 300 seconds, as shown in Figure 1. In Figure 1, the horizontal axis represents wavelength (nm) and the vertical axis represents reflection intensity.

[0040] Figure 2 shows the spectrum of reflected light from tea leaves with steaming time as a parameter, with the horizontal axis representing wavelength (nm) and the vertical axis representing reflection intensity. In the following explanation, steaming time is used as the variable corresponding to the degree of steaming. A closer look at the spectrum in Figure 2 reveals several wavelength bands that appear to depend on the steaming time. That is, in the vicinity of a wavelength of 540 nm, the reflection intensity decreases as the steaming time increases, while in the vicinity of 585 nm, it increases. On the other hand, in the wavelength range of 420 nm to 490 nm, there is almost no change in the reflection spectrum even when the steaming time is changed.

[0041] Table 1 shows a portion of the correlation coefficient matrix obtained for the correlation between the reflected light intensity of tea leaves and steaming time across the entire wavelength range, and Figure 3 plots these correlation coefficients. In Figure 3, the horizontal axis represents wavelength (nm) and the vertical axis represents reflectance. The minimum correlation coefficient was -0.28251 and the maximum was 0.739061. [Table 1]

[0042] The intensity of reflected light from tea leaves has a certain degree of correlation with steaming time (steaming level), but it is found that there is a particularly high correlation with steaming time (steaming level) in a specific wavelength range. Specifically, a strong positive correlation is observed in wavelengths around 590 nm, 640 nm, 680 nm, and 720 nm, and a negative correlation is observed in the vicinity of 540 nm. In the present invention, this correlation is utilized to estimate steaming time (steaming level) by measuring the reflection intensity of light reflected from tea leaves at a specific wavelength.

[0043] Hereinafter, embodiments of the tea leaf steaming degree estimation device of the present invention will be described in detail. However, the embodiments described below are merely schematic and illustrative examples for understanding the present invention, and the appearance and dimensional ratios may differ from the actual product. Furthermore, the present invention is not limited to these embodiments.

[0044] (First embodiment) FIG. 4 shows a first embodiment of the present invention, a tea leaf steaming degree estimation device using a neural network-based prediction model (hereinafter referred to as an AI prediction model). The steaming degree estimation device in this embodiment may be located anywhere in the tea leaf manufacturing equipment. However, the second measurement means for measuring the light intensity of a specific wavelength of reflected light from the tea leaves is preferably located at a position where the reflected light intensity of the tea leaves after passing through the steamer of the manufacturing equipment can be measured, such as the steamer outlet or the conveying section between the steamer and the subsequent process. If a cooling treatment section is included in the process subsequent to the steamer, the target tea leaves may be those that have undergone the cooling treatment. Using tea leaves with reduced surface moisture after cooling as the target tea leaves further improves measurement accuracy. Furthermore, the steaming degree determined by the steaming degree estimation device of this embodiment is returned to the steamer, enabling a reliable and appropriate steaming treatment to be performed.

[0045] As shown in FIG. 4, the steaming degree estimation device according to this embodiment includes a first measuring means 10 for measuring the light intensity of reflected light at a wavelength (specific wavelength) that shows a characteristic reflection intensity when light in the visible range (380 to 780 nm) is irradiated onto tea leaves having a known steaming time (steaming degree), a prediction model creation means 11 connected to the first measuring means 10 for creating an AI prediction model from the light intensities of reflected light at specific wavelengths of multiple tea leaves obtained by measurement by the first measuring means 10 and the known steaming times (steaming degrees) of the multiple tea leaves, and a prediction model creation means 12 for creating an AI prediction model from the light intensities of reflected light at specific wavelengths of multiple tea leaves obtained by measurement by the first measuring means 10 and the known steaming times (steaming degrees) of the multiple tea leaves. The apparatus is provided with a storage means 12 for storing the AI ​​prediction model created by the measurement model creation means 11, a second measurement means 13 for measuring the light intensity of reflected light of a specific wavelength when light in the visible range (380 to 780 nm) is irradiated onto target tea leaves for which the steaming time (steaming degree) is to be estimated, and a steaming degree estimation means 14 for reading out the AI ​​prediction model stored in the storage means 12 and using this AI prediction model to determine the steaming time (steaming degree) corresponding to the light intensity of reflected light of the specific wavelength of the tea leaves obtained by the measurement by the second measurement means 13.

[0046] In this embodiment, the first measurement means 10 and the second measurement means 13 are equipped with a multi-wavelength spectroscopic camera, such as a hyperspectral camera or multispectral camera, which separates light into wavelengths and captures the images. Although not shown, the system is configured, for example, to place the tea leaves to be measured on a movable stage, irradiate the tea leaves with wide-band light from a light source (e.g., a halogen light source or an LED light source), and measure the light intensity of the light reflected from the tea leaves using the multi-wavelength spectroscopic camera. As the movable stage moves in the scanning direction, the placed tea leaves are line-scanned by the multi-wavelength spectroscopic camera, and the reflected light intensity of the tea leaves at each wavelength is measured. As is well known, a multi-wavelength spectroscopic camera separates a two-dimensional planar image in the X and Y directions and outputs a data cube with layers for each wavelength, making it easy to obtain the spectrum of reflected light for each pixel. Therefore, it is possible to directly measure the light intensity of reflected light at multiple specific wavelengths from the tea leaves. It should be noted that the first measurement means 10 and the second measurement means 13 may be configured to use a single multi-wavelength spectroscopic camera instead of two separate multi-wavelength spectroscopic cameras.

[0047] In this embodiment, the predictive model creation means 11 is a means for creating an AI predictive model using a neural network that creates a machine learning model through deep learning using the light intensities of multiple reflected light beams at multiple specific wavelengths measured by the first measuring means 10 as explanatory variables and the known steaming times (steaming degrees) of multiple tea leaf samples as objective variables.

[0048] FIG. 5 shows an example of an AI prediction model created by the prediction model creation means 11 in this embodiment.

[0049] The AI ​​prediction model shown in Figure 5 is composed of an input layer 11a, two intermediate layers 11b and 11c that use a ReLU function (ramp function) as an activation function, and an output layer 11d that uses a Linear function as an activation function, with 3000 epochs and a mini-batch size of 50. This AI prediction model is an initial model created by the inventors in Python, and it is desirable to further improve it in the future.

[0050] The AI ​​prediction model of this embodiment uses 12 input signals (explanatory variables) consisting of 11 reflected light intensities measured at multiple specific wavelengths with characteristic reflection intensities from the reflected light spectrum in the visible and near-infrared range (actually, the 11 measured reflected light intensities divided by the reflected light intensity at 539.16 nm (wavelength number B080) as two-wavelength ratios), plus one standardized reflected light intensity at wavelength number B080. The output signal (objective variable) corresponds to the steaming degree and the steaming time (seconds). The multiple wavelengths with characteristic reflection intensities were determined based on the correlation coefficient between the objective variable, steaming time, and the reflected light intensity of 251 wavelengths across all wavelengths (400 to 1100 nm). These wavelengths were determined as having high and low correlations (correlation close to 0), for a total of 11 wavelengths. The specific wavelengths selected are 528 nm to 550 nm corresponding to pheophytin, 580 nm to 660 nm corresponding to chlorophyll, and / or 670 nm to 730 nm corresponding to chlorophyll. Examples of the correspondence between wavelength numbers and wavelengths include B001: 376.32 nm, B013: 400.73 nm, B085: 549.63 nm, B145: 676.87 nm, B181: 754.6 nm, and B289: 993.96 nm.

[0051] This AI prediction model is actually a deep learning model that divides 198 (n=198) pieces of learning data into 158 (n=158) pieces of training sample data for model creation and 40 (n=40) pieces of validation sample data for validation, and repeats learning. To confirm the generalization performance of the model, it is applied to three groups of data, including two other types of sample data that were not used to create the model. In other words, it is applied to 40 (n=40) pieces of validation sample data for validation, 22 (n=22) pieces of unknown sample data (michi data) for evaluating the model's predictive accuracy, and 60 (n=60) pieces of completely unknown sample data (shugo data) that are a collection of second-grade tea, and the evaluation index, R 2The coefficient of determination (COE), MSE (Mean Squared Error), and RMSE (Root Mean Squared Error) were calculated. The training sample data, validation sample data, and unknown sample data were measured using steamed leaves (single leaves) of the first harvest of tea, while the completely unknown sample data were measured using steamed leaves (aggregate leaves) of the second harvest of tea, which was harvested at a different time.

[0052] The AI ​​prediction model thus created by the prediction model creation means 11 is stored in the storage means 12.

[0053] The steaming degree estimation means 14 is configured to measure the light intensity of reflected light of a specific wavelength when light in the visible range is irradiated onto the target tea leaves whose steaming degree (steaming time) is to be estimated, using the second measurement means 13, read out the AI ​​prediction model stored in the memory means 12, and use this AI prediction model to determine the steaming time corresponding to the light intensity of reflected light of the specific wavelength of the tea leaves obtained by the measurement.

[0054] Next, the operation of the tea leaf steaming degree estimation device in this embodiment will be described. Figure 6 explains the operation of this steaming degree estimation device in each step.

[0055] First, in the first measurement step, light is irradiated onto a plurality of tea leaf samples each having a known steaming time, and the light intensities of reflected light at a plurality of specific wavelengths are measured (step S1).

[0056] Next, as a prediction model creation process, an AI prediction model is created that represents the correspondence between the light intensity of reflected light at multiple specific wavelengths measured in the first measurement process and the known steaming times for multiple tea leaf samples (step S2).

[0057] Next, as a storage step, the AI ​​prediction model created in the prediction model creation step is stored in the storage means 12 (step S3).

[0058] Thereafter, in the second measurement step, light is irradiated onto the target tea leaves whose steaming degree is to be estimated, and the light intensity of the reflected light at a specific wavelength is measured (step S4).

[0059] Next, as a steaming degree estimation step, the steaming time of the target tea leaves corresponding to the light intensity of the reflected light measured in the second measurement step is obtained from the AI ​​prediction model stored in the storage step (step S5).

[0060] Next, the evaluation results of the steaming time estimated by the steaming degree estimation device of this embodiment will be described. This evaluation is based on the R 2 (coefficient of determination), MSE (mean square error), and RMSE (root mean square error) were used.

[0061] Figure 7 shows the relationship between the steaming time estimated by the AI ​​prediction model and the actual steaming time for validation sample data (n=40). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this validation sample data, the relationship is y=0.978x+10.382, R 2 =0.9663, MSE=339.4832, RMSE=18.42507 were obtained.

[0062] Figure 8 shows the relationship between the steaming time estimated by the AI ​​prediction model and the actual measured steaming time for n=22 unknown sample data. The horizontal axis represents the actual measured steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relationship is y=1.0103x+7.2948, R 2 =0.941, MSE=587.60, RMSE=24.2 were obtained.

[0063] Figure 9 shows the relationship between the steaming time estimated by the AI ​​prediction model and the actual steaming time for the completely unknown sample data, which is a collection of second-grade tea leaves (n=60). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relationship formula is y=1.056x+23.28, R 2 =0.9186, MSE=1947.68, RMSE=44.13 were obtained.

[0064] The validation sample data and unknown sample data are calculated using the coefficient of determination R 2 The mean square error (RMSE) was high at 0.9663 and 0.941, and the root mean square error (RMSE) was small at 18.42507 and 24.2, resulting in good estimates. Furthermore, relatively good estimates were obtained even for completely unknown sample data. In other words, by using a machine learning model created through deep learning as a predictive model, it was found that the steaming time (steaming level) of steamed leaves can be estimated with high accuracy.

[0065] As described above, according to this embodiment, a machine learning model created by deep learning using reflected light intensity at multiple specific wavelengths with high correlation as explanatory variables and steaming time (corresponding to steaming level) as the objective variable is used as a predictive model. This allows the steaming time (steaming level) of target tea leaves to be estimated with high accuracy and reliability from the reflected light intensity of the target tea leaves. Furthermore, because the reflected light intensity at specific wavelengths is used as the explanatory variable, it is not necessary to measure the reflected light intensity at all wavelengths using an expensive multi-wavelength spectroscopic camera such as a hyperspectral camera or multispectral camera. Reflected light intensity can also be measured at low cost using a spectrometer that can measure the intensity of light separated by wavelength. An RGB camera or smartphone camera can also be used as long as it can measure the light intensity of reflected light at specific wavelengths.

[0066] As a modification of this embodiment, instead of using the reflected light spectrum as is, it is possible to use its second derivative spectrum as an explanatory variable. The second derivative spectrum allows clear confirmation of changes even in areas where the peaks and valleys are unclear in a normal reflected light spectrum, where there is little unevenness. This makes it easy to identify the wavelengths of characteristic reflection intensities.

[0067] As another modification of this embodiment, since there is some correlation between the reflected light intensity at multiple wavelengths and the steaming time (steaming degree), an AI prediction model created by machine learning can be used, with the reflected light intensity at multiple wavelengths (not a specific wavelength) as the explanatory variable and the steaming time as the objective variable. This makes it possible to estimate the steaming degree of the target tea leaves with a certain degree of reliability from the reflected light intensity of the target tea leaves.

[0068] (Second embodiment) FIG. 10 shows a second embodiment of the present invention, a tea leaf steaming degree estimation device using a multiple linear regression (MLR) prediction model. While the steaming degree estimation device in this embodiment may be located anywhere in the tea leaf manufacturing apparatus, the second measurement means for measuring the light intensity of reflected light at a specific wavelength from the tea leaves is preferably located at a position where the reflected light intensity of the tea leaves after passing through the steamer of the manufacturing apparatus can be measured, such as the steamer outlet or the conveying section between the steamer and the subsequent process. If a cooling treatment section is included in the process subsequent to the steamer, the tea leaves after the cooling treatment may be used as the target tea leaves. Using tea leaves with reduced surface moisture after cooling as the target tea leaves further improves measurement accuracy. Furthermore, the steaming degree determined by the steaming degree estimation device of this embodiment is returned to the steamer, enabling a reliable and appropriate steaming treatment to be performed.

[0069] As shown in FIG. 10, the steaming degree estimation device according to this embodiment includes a first measurement means 110 for measuring the light intensity of reflected light at a wavelength (specific wavelength) that shows a characteristic reflection intensity when light in the visible range (380 to 780 nm) is irradiated onto tea leaves having a known steaming time (steaming degree), a prediction model creation means 111 connected to the first measurement means 110 for creating an MLR prediction model from the light intensities of reflected light at specific wavelengths of multiple tea leaves obtained by measurement by the first measurement means 110 and the known steaming times (steaming degrees) of the multiple tea leaves, and a prediction The device is equipped with a storage means 112 that stores the MLR prediction model created by the model creation means 111, a second measurement means 113 that measures the light intensity of reflected light of a specific wavelength when light in the visible range (380 to 780 nm) is irradiated onto target tea leaves whose steaming time (steaming degree) is to be estimated, and a steaming degree estimation means 114 that reads out the MLR prediction model stored in the storage means 112 and uses this MLR prediction model to determine the steaming time (steaming degree) corresponding to the light intensity of reflected light of the specific wavelength of the tea leaves obtained by the measurement by the second measurement means 113.

[0070] In this embodiment, the first measurement means 110 and the second measurement means 113 are equipped with a multi-wavelength spectroscopic camera, such as a hyperspectral camera or multispectral camera, which separates light into wavelengths and captures the images. Although not shown, the system is configured, for example, to place the tea leaves to be measured on a movable stage, irradiate the tea leaves with wide-band light from a light source (e.g., a halogen light source or an LED light source), and measure the light intensity of the light reflected from the tea leaves using the multi-wavelength spectroscopic camera. As the movable stage moves in the scanning direction, the placed tea leaves are line-scanned by the multi-wavelength spectroscopic camera, and the reflected light intensity of the tea leaves at each wavelength is measured. As is well known, a multi-wavelength spectroscopic camera separates a two-dimensional planar image in the X and Y directions and outputs a data cube with layers for each wavelength, making it easy to obtain the spectrum of reflected light for each pixel. Therefore, it is possible to directly measure the light intensity of reflected light at multiple specific wavelengths from the tea leaves. It should be noted that the first measurement means 110 and the second measurement means 113 may be configured to use a single multi-wavelength spectroscopic camera instead of two separate multi-wavelength spectroscopic cameras. Furthermore, since this embodiment uses a small number of explanatory variables, an RGB camera, a smartphone camera, or a spectrometer capable of measuring the intensity of light dispersed for each wavelength may be used instead of the multi-wavelength spectroscopic camera. An RGB camera or a smartphone camera may also be used as long as it can measure the light intensity of reflected light at a specific wavelength.

[0071] In this embodiment, the predictive model creation means 111 is a predictive model creation means that creates a multiple regression equation with the light intensities of multiple reflected light beams at multiple specific wavelengths measured by the first measurement means 110 as explanatory variables and the known steaming time (steaming degree) of the tea leaf sample as the objective variable.

[0072] FIG. 11 shows an example of an MLR prediction model created by the prediction model creation means 111 in this embodiment and an explanatory variable narrowing means for narrowing down the explanatory variables.

[0073] The prediction model creation means 111 shown in FIG. 11 is composed of an explanatory variable narrowing means 111a that performs PLS (Partial Least Squares) analysis (GA-PLS) using a genetic algorithm (GA) to narrow down effective explanatory variables, and a multiple regression (MLR) model 111b consisting of a multiple regression prediction formula in which the reflected light intensity of a small number of wavelengths obtained by narrowing down the explanatory variables is used as an explanatory variable.

[0074] The explanatory variable narrowing means 111a uses 11 reflected light intensities measured at multiple wavelengths with characteristic reflection intensities from the reflected light spectrum in the visible and near-infrared regions (actually, the 11-item two-wavelength ratio obtained by dividing the measured reflected light intensities at 11 items by the reflected light intensity at 539.16 nm (wavelength number B080)), plus one term standardizing the reflected light intensity at wavelength number B080, as input signals (explanatory variables), and uses steaming degree corresponding to steaming time (seconds) as an output signal (objective variable), and selects and narrows down important explanatory variables using GA-PLS analysis. Specifically, from among multiple candidate combinations of explanatory variables, three wavelengths were selected: 549.64 nm (wavelength number B085), 587.5 nm (wavelength number B103), and 629.88 nm (wavelength number B123), which had a small number of terms and were comparable in evaluation index to models with a large number of explanatory variables. The items actually selected were three items: B085 / B080, B103 / B080, and B123 / B080, which were standardized by the reflected light intensity of wavelength number B080.

[0075] The MLR prediction model 111b is a multiple regression prediction formula y = -606.719x1 + 990.916x2 - 590.927x3 + 288.6753 created by analyzing these three items using MLR in the R language, where y represents the estimated steaming time (seconds), x1 represents the reflected light intensity (rate) of B085 / B080, x2 represents the reflected light intensity (rate) of B103 / B080, and x3 represents the reflected light intensity (rate) of B123 / B080.

[0076] This MLR prediction model is actually an MLR prediction model (multiple regression prediction formula) that divides 198 (n=198) pieces of learning data into 158 (n=158) pieces of training sample data for model creation and 40 (n=40) pieces of validation sample data for validation, and performs MLR analysis. In order to confirm the generalization performance of the model, the MLR prediction model is applied to three groups of data, including two other types of sample data that were not used to create the model. In other words, it is applied to 40 (n=40) pieces of validation sample data, 22 (n=22) pieces of unknown sample data (michi data), and 60 (n=60) pieces of completely unknown sample data (shugo data), which is a collection of second-grade tea, and the evaluation index, R 2 The coefficient of determination (COE), MSE (Mean Squared Error), and RMSE (Root Mean Squared Error) were calculated. The training sample data, validation sample data, and unknown sample data were measured using steamed leaves (single leaves) of the first harvest of tea, while the completely unknown sample data were measured using steamed leaves (aggregate leaves) of the second harvest of tea, which was harvested at a different time.

[0077] The MLR prediction model thus created by the prediction model creation means 111 is stored in the storage means 112.

[0078] The steaming degree estimation means 114 is configured to measure the light intensity of reflected light of a specific wavelength when light in the visible range is irradiated onto the target tea leaves whose steaming degree (steaming time) is to be estimated, using the second measurement means 113, read out the MLR prediction model (multiple regression prediction formula) stored in the memory means 112, and use this multiple regression prediction formula to determine the steaming time corresponding to the light intensity of reflected light of the specific wavelength of the tea leaves obtained by the measurement.

[0079] Next, the operation of the tea leaf steaming degree estimation device in this embodiment will be described. Figure 12 explains the operation of each step of this steaming degree estimation device.

[0080] First, the explanatory variables are narrowed down by GA-PLS analysis (step S11). Here, the 12 explanatory variables are narrowed down to three items that are promising combinations of variables.

[0081] Next, as a first measurement step, light is irradiated onto a plurality of tea leaf samples each having a known steaming time, and the light intensities of reflected light at a plurality of specific wavelengths are measured (step S12).

[0082] Next, as a prediction model creation process, an MLR prediction model (multiple regression prediction formula) is created to perform MLR analysis based on the correspondence between the light intensity of reflected light at multiple specific wavelengths measured in the first measurement process and the known steaming times for multiple tea leaf samples (step S13).

[0083] Next, as a storage step, the MLR prediction model created in the prediction model creation step is stored in the storage means 112 (step S14).

[0084] Thereafter, in the second measurement step, light is irradiated onto the target tea leaves whose steaming degree is to be estimated, and the light intensity of the reflected light at a specific wavelength is measured (step S15).

[0085] Next, as a steaming degree estimation step, the MLR prediction model (multiple regression prediction formula) stored in the storage step is used to determine the steaming time of the target tea leaves corresponding to the light intensity of the reflected light measured in the second measurement step (step S16).

[0086] Next, the evaluation results of the steaming time estimated by the steaming degree estimation device of this embodiment will be described. This evaluation is based on the R 2 (coefficient of determination), MSE (mean square error), and RMSE (root mean square error) were used.

[0087] Figure 13 shows the relationship between the steaming time estimated by the MLR prediction model and the actual steaming time for n=158 training sample data. The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this training sample data, the relationship is y=0.97x+4.3981, R 2 =0.97, MSE=276.5275, RMSE=16.63 were obtained.

[0088] Figure 14 shows the relationship between the steaming time estimated by the MLR prediction model and the actual steaming time for validation sample data (n=40). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this validation sample data, the relationship is y=0.9813x+1.056, R 2 =0.9628, MSE=320.236, RMSE=17.90 were obtained.

[0089] Figure 15 shows the relationship between the steaming time estimated by the MLR prediction model and the actual measured steaming time for n=22 unknown sample data. The horizontal axis represents the actual measured steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relational expression is y=0.9729x+9.884, R 2 =0.9467, MSE=464.9247, RMSE=21.56 were obtained.

[0090] Figure 16 shows the relationship between the steaming time estimated by the MLR prediction model and the actual steaming time for the completely unknown sample data, which is a collection of second-grade tea leaves (n=60). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relational expression is y=0.9726x+22.339, R 2 =0.9473, MSE=858.428, RMSE=29.30 were obtained.

[0091] The results of self-application of the MLR prediction model to the sample data (training sample data) for model creation were evaluated using R. 2 =0.97, RMSE=16.63. As a result of the validation, R 2 =0.9628, MSE=320.236, RMSE=17.90, R with unknown sample data 2 =0.9467, MSE=464.9247, RMSE=21.56, R with completely unknown sample data 2 = 0.9473, MSE = 858.426, RMSE = 29.30, showing good prediction accuracy. In other words, it was found that using such an MLR prediction model as a prediction model allows for highly accurate estimation of the steaming time (steaming level) of steamed leaves. Furthermore, when comparing the results of application to validation sample data and unknown sample data for first-crop tea (single leaf) with the completely unknown sample data for the second-crop tea aggregate based on the evaluation index, it was found that the differences were not as large as in the case of the AI ​​prediction model of the first embodiment. This indicates that even with the conventional modeling method, the MLR method, if appropriate explanatory variables are selected, a model with high generalization performance can be obtained. Because the MLR prediction model is a modeling method that is expected to be put to practical use, careful consideration of input variables will become even more important in the future.

[0092] As described above, according to this embodiment, the prediction model is an MLR prediction model that uses reflected light intensities at multiple specific wavelengths with high correlation as explanatory variables and steaming time (corresponding to steaming degree) as the objective variable. Therefore, the steaming time (steaming degree) of target tea leaves can be estimated with high accuracy and reliability from the reflected light intensity of the target tea leaves. Furthermore, by narrowing down the model and selecting only valid explanatory variables to reduce the number of terms, the multiple regression prediction formula can be significantly simplified. Furthermore, compared to models without variable selection (models with many explanatory variables and many terms in the formula), the prediction accuracy is comparable to that of models without variable selection. Of course, the number of explanatory variables narrowed down is not limited to three terms; it can be four or more terms. Furthermore, because the reflected light intensity at three wavelengths is used, it is not necessary to measure the reflected light intensity at all wavelengths using an expensive multi-wavelength spectroscopic camera such as a hyperspectral camera or multispectral camera. Reflected light intensity can also be measured at low cost using a spectrometer that can measure the intensity of light separated by wavelength. An RGB camera or smartphone camera can also be used as long as it can measure the light intensity of reflected light at specific wavelengths.

[0093] As a modification of this embodiment, instead of using the reflected light spectrum as is, it is possible to use its second derivative spectrum as an explanatory variable. The second derivative spectrum allows clear confirmation of changes even in areas where the peaks and valleys are unclear in a normal reflected light spectrum, where there is little unevenness. This makes it easy to identify the wavelengths of characteristic reflection intensities.

[0094] As another modification of this embodiment, since there is some correlation between the reflected light intensity at multiple wavelengths and the steaming time (steaming degree), an MLR prediction model can be used that is created by MLR analysis using the reflected light intensity at multiple wavelengths other than a specific wavelength as the explanatory variable and the steaming time as the objective variable. This makes it possible to estimate the steaming degree of the target tea leaves with a certain degree of reliability from the reflected light intensity of the target tea leaves.

[0095] (Third embodiment) FIG. 17 shows a third embodiment of the present invention, a tea leaf steaming degree estimation device using a partial least squares (PLS) regression model. The steaming degree estimation device in this embodiment may be located anywhere in the tea leaf manufacturing apparatus. However, the second measurement means for measuring the light intensity of reflected light at multiple wavelengths from the tea leaves is preferably located at a position where the reflected light intensity of the tea leaves after passing through the steamer of the manufacturing apparatus can be measured, such as the steamer outlet or the conveying section between the steamer and the subsequent process. If a cooling treatment section is included in the process subsequent to the steamer, the tea leaves after the cooling treatment may be used as the target tea leaves. Using tea leaves with reduced surface moisture after cooling as the target tea leaves further improves measurement accuracy. Furthermore, the steaming degree determined by the steaming degree estimation device of this embodiment is returned to the steamer, enabling a reliable and appropriate steaming treatment to be performed.

[0096] As shown in FIG. 17, the steaming degree estimation device according to this embodiment includes a first measurement means 210 for measuring the light intensity of reflected light at a plurality of wavelengths when light in the visible region (380 to 780 nm) and the near-infrared region (wavelengths longer than 780 nm) is irradiated onto tea leaves having a known steaming time (steaming degree), a prediction model creation means 211 connected to the first measurement means 210 for creating a PLS regression model from the reflected light intensity rate obtained by dividing the light intensity of the reflected light at a plurality of wavelengths of the tea leaves measured by the first measurement means 210 by the wavelength band average value and the known steaming time (steaming degree) of the tea leaves, and a prediction model creation means 212 for creating a PLS regression model from the predicted model The apparatus is provided with a storage means 212 that stores the PLS regression model created by the model creation means 211, a second measurement means 213 that measures the light intensity of reflected light of multiple wavelengths when light in the visible range (380 to 780 nm) and near-infrared range (wavelengths longer than 780 nm) is irradiated onto target tea leaves for which the steaming time (steaming degree) is to be estimated, and a steaming degree estimation means 214 that reads out the PLS regression model stored in the storage means 212 and uses this PLS regression model to determine the steaming time (steaming degree) corresponding to the light intensity of reflected light of multiple wavelengths of the tea leaves obtained by the measurement by the second measurement means 213.

[0097] In this embodiment, the first measurement means 210 and the second measurement means 213 are equipped with a multi-wavelength spectroscopic camera, such as a hyperspectral camera or multispectral camera, which separates light into wavelengths and captures the images. Although not shown, the system is configured, for example, to place the tea leaves to be measured on a movable stage, irradiate the tea leaves with wide-band light from a light source (e.g., a halogen light source or an LED light source), and measure the light intensity of the light reflected from the tea leaves using the multi-wavelength spectroscopic camera. As the movable stage moves in the scanning direction, the placed tea leaves are line-scanned by the multi-wavelength spectroscopic camera, and the reflected light intensity of the tea leaves at each wavelength is measured. As is well known, a multi-wavelength spectroscopic camera separates a two-dimensional planar image in the X and Y directions and outputs a data cube with layers for each wavelength, making it easy to obtain the spectrum of reflected light for each pixel. Therefore, it is possible to directly measure the light intensity of reflected light at multiple wavelengths from the tea leaves. It should be noted that the first measurement means 210 and the second measurement means 213 may be configured to use a single multi-wavelength spectroscopic camera instead of two separate multi-wavelength spectroscopic cameras. Furthermore, since this embodiment uses a small number of explanatory variables, an RGB camera, a smartphone camera, or a spectrometer capable of measuring the intensity of light dispersed for each wavelength may be used instead of the multi-wavelength spectroscopic camera. An RGB camera or a smartphone camera may also be used as long as it can measure the light intensity of reflected light at a specific wavelength.

[0098] In this embodiment, the predictive model creation means 211 is a predictive model creation means that creates a PLS regression model using the reflected light intensity rate, obtained by dividing the light intensity of reflected light at multiple wavelengths (300 wavelengths) measured by the first measurement means 210 by the wavelength band average value, as an explanatory variable, and the known steaming time (steaming degree) of the tea leaf sample as a target variable.

[0099] FIG. 18 shows an example of the prediction model creation means 211 in this embodiment.

[0100] 18 is composed of division means 211a that divides the reflected light intensity at multiple wavelengths (300 wavelengths) from the first measurement means 210 by the wavelength band average value, and a PLS regression model 211b that uses the reflected light intensity rate obtained by division as the explanatory variable and the known steaming time (steaming degree) as the response variable. The partial least squares regression (PLS regression) method is a method for searching for a hyperplane with maximum variance by projecting predictor variables and observable variables into a new space when searching for a linear regression model between a response variable and an explanatory variable.

[0101] The division means 211a is a means for generating explanatory variables consisting of BandRatio-type reflected light intensities for 300 wavelengths by dividing the reflected light intensities for 300 wavelengths in the reflected light spectrum in the visible and near-infrared range measured by the first measurement means 210 by the average value of the stable wavelength band. The average reflectance spectra and second-order derivative spectra of each sample (training samples, validation samples, unknown samples, completely unknown samples) have a common region where the spectrum is flat and has little fluctuation (e.g., wavelength 466.4 nm (B045) to wavelength 474.66 nm (B049)). This wavelength band with little fluctuation is focused on as a stable wavelength band that is not easily affected by steaming time, and the average value of the five wavelengths that make up this wavelength band (wavelength band average value) is used to divide the reflected light intensity (ratio) for 300 wavelengths to generate BandRatio-type reflected light intensities, which are used as explanatory variables for the PLS regression model 211b.

[0102] The PLS regression model 211b is a PLS regression analysis model created using the R programming language. The BandRatio-type reflected light intensity over 300 wavelengths input from the division means 211a is used as the explanatory variable, and the estimated steaming time (seconds) is used as the objective variable. By using an explanatory variable obtained by dividing the difference in reflected light intensity resulting from the differences between the first- and second-grade tea leaves by the average value of a flat, less-fluctuating wavelength range, prediction accuracy was significantly improved. Note that in this embodiment, during model creation, a standardization process was performed to approximate the data distribution as a normal distribution so that different variables could be compared.

[0103] Table 2 shows some of the regression coefficients in the PLS regression model of this embodiment. [Table 2]

[0104] The PLS regression model thus created by the prediction model creation means 211 is stored in the storage means 212.

[0105] The steaming degree estimation means 214 is configured to measure the light intensity of reflected light of multiple wavelengths when light in the visible and near-infrared regions is irradiated onto the target tea leaves whose steaming degree (steaming time) is to be estimated, using the second measurement means 213, read out the PLS regression model stored in the memory means 212, and use the PLS regression model to determine the steaming time corresponding to the measured reflected light intensity of the tea leaves.

[0106] Next, the operation of the tea leaf steaming degree estimation device in this embodiment will be described. Figure 19 explains the operation of each step of this steaming degree estimation device.

[0107] First, in the first measurement step, light is irradiated onto a tea leaf sample having a known steaming time, and the light intensity of reflected light at a plurality of wavelengths (300 wavelengths) is measured (step S21).

[0108] Next, the measured reflected light intensity (300 wavelengths) is divided by the average value of the wavelength region with little fluctuation (stable wavelength band) (step S22).

[0109] Next, as a prediction model creation process, a PLS regression analysis is performed based on the correspondence between the light intensity of reflected light at multiple wavelengths measured in the first measurement process and the known steaming times for multiple tea leaf samples to create a PLS regression model (step S23).

[0110] Next, as a storage step, the PLS regression model created in the prediction model creation step is stored in the storage means 212 (step S24).

[0111] Thereafter, in the second measurement step, light is irradiated onto the target tea leaves whose steaming degree is to be estimated, and the light intensity of the reflected light at a plurality of wavelengths is measured (step S25).

[0112] Next, in a steaming degree estimation step, the PLS regression model stored in the storage step is used to determine the steaming time of the target tea leaves corresponding to the light intensity of the reflected light measured in the second measurement step (step S26).

[0113] Next, the evaluation results of the steaming time estimated by the steaming degree estimation device of this embodiment will be described. This evaluation is based on the R 2 (coefficient of determination), MSE (mean square error), and RMSE (root mean square error) were used.

[0114] Figure 20 shows the relationship between the steaming time estimated by the PLS regression model and the actual steaming time for the training sample data (n=158). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this training sample data, the relationship is y=0.9495x+7.0845, R 2 =0.9495, MSE=437.306, RMSE=20.91 were obtained.

[0115] Figure 21 shows the relationship between the steaming time estimated by the PLS regression model and the actual steaming time for validation sample data (n=40). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this validation sample data, the relationship is y=0.9551x+10.172, R 2 =0.9279, MSE=650.3034, RMSE=25.50 were obtained.

[0116] Figure 22 shows the relationship between the steaming time estimated by the PLS regression model and the actual measured steaming time for n=22 unknown sample data. The horizontal axis represents the actual measured steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relational expression is y=0.9769x+8.0855, R 2 =0.9243, MSE=648.4464, RMSE=25.46461 were obtained.

[0117] Figure 23 shows the relationship between the steaming time estimated by the PLS regression model and the actual steaming time for the completely unknown sample data, which is a collection of second-grade tea leaves (n=60). The horizontal axis represents the actual steaming time (seconds), and the vertical axis represents the estimated steaming time (seconds). For this unknown sample data, the relational expression is y=0.949x+5.2802, R 2 =0.9279, MSE=852.1369, RMSE=29.19138 were obtained.

[0118] As you can see from the validation results, the validation sample data is 2 =0.9279, MSE=650.3034, RMSE=25.50, R with unknown sample data 2 =0.9243, MSE=648.4464, RMSE=25.46461, R with completely unknown sample data 2 = 0.9279, MSE = 852.1369, RMSE = 29.19138, and all prediction accuracies were good. In other words, it was found that by using such a PLS regression model as a prediction model, the steaming time (steaming degree) of steamed leaves can be estimated with high accuracy. When PLS regression analysis was performed using the reflected light intensity measured by the first measurement means 210 as an explanatory variable without dividing it by the average value of the stable wavelength band, the MSE (mean square error) and RMSE (root mean square error) of the completely unknown sample data, which is a collection of second-grade tea leaves, were R 2 = 0.9279, MSE = 14575.32, RMSE = 120.73, and the generalization performance was very poor. This is thought to be due to the upward and downward shift in reflected light intensity resulting from the difference between the first and second harvests.

[0119] As described above, according to this embodiment, the predicted model is a PLS regression model in which the reflected light intensity obtained by dividing the measured reflected light intensity at multiple wavelengths by the average stable wavelength band is used as the explanatory variable, and the steaming time (corresponding to the steaming degree) is used as the objective variable, so the steaming time (steaming degree) of the target tea leaves can be estimated with high accuracy and reliability from the reflected light intensity of the target tea leaves. The difference in reflected light intensity between the first- and second-harvest teas, which are harvested at different times, is accounted for by the reflected light intensity obtained by dividing the average stable wavelength band, which has a flat and little fluctuation, as the explanatory variable, greatly improving the prediction accuracy.

[0120] As a modification of this embodiment, instead of using the reflected light spectrum as is, it is possible to use its second derivative spectrum as an explanatory variable. The second derivative spectrum allows clear confirmation of changes even in areas where the peaks and valleys are unclear in a normal reflected light spectrum, where there is little unevenness. This makes it easy to identify the wavelengths of characteristic reflection intensities.

[0121] In another modification of this embodiment, a PLS regression model may be used in which the reflected light intensity at multiple specific wavelengths is used as the explanatory variable and the steaming time is used as the response variable, thereby enabling the steaming level of the target tea leaves to be reliably estimated from the reflected light intensity of the target tea leaves.

[0122] The steaming degree estimation device and steaming degree estimation method for tea leaves of the present invention have been described above using the first to third embodiments and modified aspects, and the steaming degree obtained by these steaming degree estimation devices and steaming degree estimation methods is sent back to the steamer in the tea leaf manufacturing apparatus and used to control the steaming operation. Explaining the configuration of the tea leaf manufacturing apparatus, for example, an apparatus for manufacturing crude tea is merely one example, and includes a steaming treatment section that controls the steaming operation of the steamer using the steaming degree estimated by the present invention, a cooling treatment section that quickly cools the temperature of the steamed leaves to about room temperature, a leaf beating treatment section that blows dry hot air into the cooled tea leaves to shake them, a rough rolling treatment section that blows dry hot air into the tea leaves to apply moderate friction and pressure and roll them in order to soften them and reduce their internal moisture, and a rolling treatment section in the rough rolling treatment section. The mill comprises a rolling and twisting section where the tea leaves are pressed together and rolled to compensate for insufficient rolling, destroy the structure of the tea leaves to facilitate the infusion of their ingredients, and ensure uniform moisture content, a medium rolling section where the tea leaves are rolled and pressed while hot dry air is blown in to loosen the tea leaves into a twisted shape and dry them so that they can be easily shaped in the next fine rolling section, a fine rolling section where the tea leaves are rolled in only one direction while removing moisture from inside the tea leaves and drying them to give them a thin, elongated shape, and a drying section where the tea leaves are dried to reduce their moisture content to about 5%. The steaming section, cooling section, leaf beating section, coarse rolling section, rolling and twisting section, medium rolling section, fine rolling section, and drying section are all commercially available and of known construction, so detailed explanations will be omitted.

[0123] The above-described embodiments and modifications are merely illustrative of the present invention and are not intended to limit the scope of the present invention, which can be embodied in various other modified and altered forms. Therefore, the scope of the present invention is defined only by the claims and their equivalents. [Explanation of symbols]

[0124] 10, 110, 210 First measuring means 11, 111, 211 Predictive model creation methods 11a input layer, 11b, 11c middle tier 11d Output layer 12, 112, 212 Memory means 13, 113, 213 Second measurement means 14, 114, 214 Steaming degree estimation method 111a Explanatory variable narrowing method 111b MLR prediction model 211a Division means 211b PLS regression model

Claims

1. a first measuring means for irradiating a tea leaf sample having a known steaming degree with light in the visible and near-infrared regions and measuring the light intensity of reflected light at a plurality of wavelengths; a prediction model creation means for creating a steamness prediction model that represents a correspondence relationship between the light intensity of the reflected light at the plurality of wavelengths measured by the first measurement means and the known steamness of the tea leaf sample; A storage means for storing the steamness prediction model created by the prediction model creation means; a second measuring means for irradiating light in the visible and near-infrared regions onto the target tea leaves whose steaming degree is to be estimated and measuring the light intensity of the reflected light at the plurality of wavelengths; a steaming degree estimation means for estimating the steaming degree of the target tea leaves corresponding to the light intensity of the reflected light measured by the second measurement means from the steaming degree prediction model stored in the storage means; A tea leaf steaming degree estimation device comprising:

2. a first measuring means for irradiating a plurality of tea leaf samples each having a known steaming degree with light in the visible region and measuring the light intensity of reflected light at a plurality of specific wavelengths; a prediction model creation means for creating a steamness prediction model that represents a correspondence relationship between the light intensities of the reflected light at the plurality of specific wavelengths measured by the first measurement means and the known steamness levels of the plurality of tea leaf samples; A storage means for storing the steamness prediction model created by the prediction model creation means; a second measuring means for irradiating light in the visible region onto the target tea leaves whose steaming degree is to be estimated and measuring the light intensity of the reflected light at the specific wavelength; a steaming degree estimation means for estimating the steaming degree of the target tea leaves corresponding to the light intensity of the reflected light measured by the second measurement means from the steaming degree prediction model stored in the storage means; A tea leaf steaming degree estimation device comprising:

3. The tea leaf steaming degree estimation device according to claim 2, characterized in that the prediction model creation means is a means for creating a machine learning model created by deep learning, in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured by the first measurement means are used as explanatory variables and the known steaming degrees of the multiple tea leaf samples are used as objective variables.

4. The device for estimating a steaming degree of tea leaves according to claim 2, characterized in that the prediction model creation means is means for creating a multiple regression model having a multiple regression prediction formula in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured by the first measurement means are used as explanatory variables and the known steaming degree of the tea leaf sample is used as a response variable.

5. 5. The tea leaf steaming degree estimating device according to claim 4, wherein the prediction model creating means comprises narrowing-down means for selecting a small number of wavelengths from the plurality of specific wavelengths to narrow down effective explanatory variables.

6. 2. The device for estimating a steaming degree of tea leaves according to claim 1, wherein the prediction model creation means is means for creating a partial least squares regression model having a partial least squares regression prediction formula in which the light intensity of the reflected light at the plurality of wavelengths measured by the first measurement means is an explanatory variable and the known steaming degree of the tea leaf sample is a response variable.

7. The tea leaf steaming degree estimating device according to claim 6, characterized in that the prediction model creating means includes a division means for dividing the light intensity of the reflected light at the plurality of wavelengths measured by the first measuring means by the wavelength band average value to obtain explanatory variables.

8. 3. The tea leaf steaming degree estimating device according to claim 2, wherein the specific wavelengths include wavelengths of 528 nm to 550 nm.

9. 3. The tea leaf steaming degree estimating device according to claim 2, wherein the specific wavelengths include wavelengths of 580 nm to 660 nm.

10. 3. The tea leaf steaming degree estimating device according to claim 2, wherein the specific wavelengths include wavelengths of 670 nm to 730 nm.

11. 3. The tea leaf steaming degree estimating device according to claim 2, wherein the first measuring means and the second measuring means are light intensity measuring means equipped with a multi-wavelength spectroscopic camera.

12. The tea leaf steaming degree estimation device according to claim 2, characterized in that the second measurement means is configured to measure the light intensity of reflected light at the specific wavelength of the target tea leaves after steaming treatment using a steamer.

13. 3. The device for estimating a steaming degree of tea leaves according to claim 2, wherein the steaming degree estimating means is configured to transmit the determined steaming degree to a tea leaf steamer.

14. a first measuring step of irradiating a tea leaf sample having a known steaming degree with light in the visible and near-infrared regions and measuring the light intensity of reflected light at a plurality of wavelengths; a prediction model creation step of creating a steamness prediction model that represents a correspondence relationship between the light intensities of the reflected light at the plurality of wavelengths measured in the first measurement step and the known steamness levels of the plurality of tea leaf samples; a storage step of storing the steaming degree prediction model created in the prediction model creation step; a second measuring step of irradiating light in the visible and near-infrared regions onto the target tea leaves whose steaming degree is to be estimated and measuring the light intensity of the reflected light at the plurality of wavelengths; a steamness estimation step of determining the steamness of the target tea leaves corresponding to the light intensity of the reflected light measured in the second measurement step from the steamness prediction model stored in the storage step; A method for estimating the steaming degree of tea leaves, comprising:

15. a first measuring step of irradiating light in the visible region onto a plurality of tea leaf samples each having a known steaming degree and measuring the light intensity of reflected light at a plurality of specific wavelengths; a prediction model creation step of creating a steamness prediction model that represents a correspondence relationship between the light intensities of the reflected light at the plurality of specific wavelengths measured in the first measurement step and the known steamness levels of the plurality of tea leaf samples; a storage step of storing the steaming degree prediction model created in the prediction model creation step; a second measuring step of irradiating light in the visible region onto the target tea leaves whose steaming degree is to be estimated and measuring the light intensity of the reflected light at the specific wavelength; a steamness estimation step of determining the steamness of the target tea leaves corresponding to the light intensity of the reflected light measured in the second measurement step from the steamness prediction model stored in the storage step; A method for estimating the steaming degree of tea leaves, comprising:

16. The method for estimating the steaming degree of tea leaves according to claim 15, characterized in that the prediction model creation step is a step of creating a machine learning model created by deep learning, in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured in the first measurement step are used as explanatory variables and the known steaming degrees of the multiple tea leaf samples are used as objective variables.

17. The method for estimating the steaming degree of tea leaves according to claim 15, characterized in that the prediction model creation step is a step of creating a multiple regression model equipped with a multiple regression prediction formula in which the light intensities of the multiple reflected light beams at the multiple specific wavelengths measured in the first measurement step are used as explanatory variables and the known steaming degree of the tea leaf sample is used as a target variable.

18. The method for estimating the steaming degree of tea leaves according to claim 17, characterized in that the prediction model creation step includes a narrowing down step of selecting a small number of wavelengths from the plurality of specific wavelengths to narrow down effective explanatory variables.

19. 15. The method for estimating the steaming degree of tea leaves according to claim 14, wherein the prediction model creation step is a step of creating a partial least squares regression model having a partial least squares regression prediction formula in which the light intensity of the reflected light at the plurality of wavelengths measured in the first measurement step is used as an explanatory variable and the known steaming degree of the tea leaf sample is used as a target variable.

20. 20. The method for estimating the steaming degree of tea leaves according to claim 19, characterized in that the prediction model creation step includes a division step of dividing the light intensity of the reflected light at the plurality of wavelengths measured in the first measurement step by the wavelength band average value to obtain explanatory variables.

21. 16. The method for estimating the steaming degree of tea leaves according to claim 15, wherein the second measuring step measures the light intensity of reflected light at the specific wavelength of the target tea leaves after steaming treatment using a steamer.

22. 16. The method for estimating the steaming degree of tea leaves according to claim 15, wherein the steaming degree calculated in the steaming degree estimating step is transmitted to a tea leaf steamer.

23. A method for producing tea leaves, characterized in that the steaming degree of tea leaves is estimated by the method for estimating the steaming degree of tea leaves according to any one of claims 14 to 22.

Images