Evapotranspiration learning data creation device, machine learning device, estimation device, method, and program

Abstract

To provide a learning data creation device, a machine learning device, an estimation device, a learning data creation method, and a machine learning method that quantitatively grasp a transpiration rate of a plant.SOLUTION: In a transpiration rate estimation support system, a learning data creation device 1 includes: a single leaf transpiration rate calculation unit 11 that calculates, on the basis of environmental data 41 in a farm field, a single leaf transpiration rate; an index calculation unit 12 that calculates, on the basis of an image 42 including a leaf of a plant cultivated in the farm field, an index related to the leaf; and a learning data creation unit 13 that creates learning data 4 by associating a plant transpiration rate 43 measured in the farm field with the single leaf transpiration rate and the index related to the leaf.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a technique for estimating the transpiration rate of plants.

Background Art

[0002] The growth of plants proceeds through carbon assimilation by photosynthesis. Also, the water balance in plants governs the maintenance of form, the elongation growth of cells, the opening and closing movements of stomata, etc. In particular, water deficiency in leaves leads to stomatal closure and insufficient supply of carbon dioxide into the leaves, suppressing photosynthesis and greatly affecting plant production. The photosynthesis rate indicating the degree to which the photosynthesis reaction is proceeding in plants and the transpiration rate indicating the degree to which water is transpired from the leaves of plants are known as factors that greatly influence the growth status and yield of plants. Continuously and quantitatively grasping the photosynthesis rate and transpiration rate of crops over a long period from when the crops are planted in the field until harvest is important for the farming sites aiming to improve agricultural efficiency.

[0003] In farming sites, the transpiration rate of crops is measured by installing a transparent open-type chamber surrounding the crops in the field. The crops planted in the field are surrounded by a transparent open-type chamber for measurement, and outside air is ventilated into the chamber. The H2O concentration in the chamber changes as water is transpired from the leaves of the crops while the outside air passes through the chamber. The transpiration rate of the crops can be measured using a gas analyzer and a flow meter based on the concentration difference of H2O between the intake port and the exhaust port provided in the chamber. Regarding the photosynthesis rate of crops as well as the transpiration rate, it can be measured based on the concentration difference of CO2 between the intake port and the exhaust port.

[0004] In recent years, efforts have also been made to introduce information and communication technology (ICT) and artificial intelligence (AI) in farming sites. For example, according to the production management system described in Patent Document 1 below, it is possible to predict the yield of agricultural crops based on a model showing the photosynthesis amount of agricultural crops with respect to temperature and a model showing the growth amount of agricultural crops with respect to temperature.

Prior Art Documents

Patent Document

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] Measurement systems of the type installed in the field using an open chamber are expensive and require advanced measurement techniques, so they are mainly used for academic research purposes. It is not realistic for such measurement systems used by researchers to be installed in the farming field and operated daily, especially by small-scale producers.

[0007] The production management system described in Patent Document 1 is a system for predicting the yield of agricultural crops, not a system for predicting the transpiration rate of agricultural crops. Although the production management system of Patent Document 1 uses a model showing the photosynthesis amount of agricultural crops with respect to temperature, this model is not a model for predicting the transpiration rate.

[0008] Under such circumstances, in the farming field, especially in small-scale fields, the growth status of crops is still grasped based on the experience and intuition of producers, and efforts to measure and quantitatively grasp the transpiration rate of crops have not advanced. In order to appropriately grasp the growth status of crops and improve the efficiency of agriculture, it is required to more easily and quantitatively grasp the transpiration rate of crops in the farming field. This is the same for cut flowers as well as for agricultural crops.

[0009] An object of the present invention is to quantitatively grasp the transpiration rate of plants.

Means for Solving the Problems

[0010] The present invention for achieving the above object includes, for example, the following aspects. (Item 1) A learning data creation device for creating learning data used for learning an estimation algorithm for estimating the transpiration rate of a plant, a single-leaf transpiration rate calculation unit that calculates the single-leaf transpiration rate based on environmental data in a field, an index calculation unit that calculates an index related to the leaf based on an image including the leaf of a plant cultivated in the field, a learning data creation unit that creates learning data by associating the transpiration rate of the plant measured in the field with the single-leaf transpiration rate and the index related to the leaf, A learning data creation device comprising: (Item 2) The learning data creation device according to item 1, wherein the index calculation unit includes a leaf region extraction unit that extracts a region of the leaf in the image by image analysis using artificial intelligence. (Item 3) The learning data creation device according to any one of items 1 or 2, wherein the index related to the leaf includes at least one of a leaf area index of the leaf and a light reception efficiency index of the leaf. (Item 4) The learning data creation device according to item 3, wherein the index calculation unit includes a leaf area index calculation unit that calculates the leaf area index using a ratio occupied by a region other than the leaf in the image obtained by binarizing the image. (Item 5) The learning data creation device according to any one of items 3 or 4, wherein the index calculation unit includes a light reception efficiency index calculation unit that calculates the light reception efficiency index by integrating the relative luminance of the leaf region in the image. (Item 6) The learning data creation device according to any one of items 1 to 5, wherein the single-leaf transpiration rate calculation unit calculates the single-leaf transpiration rate based on the environmental data and a plant physiological ecology model. (Item 7) The plant physiological ecology model includes a transpiration demand model, The environmental data includes a photosynthetic photon flux density, an atmospheric temperature, an atmospheric humidity, and a leaf surface boundary layer conductance for heat transport. The individual leaf transpiration rate calculation unit calculates the individual leaf transpiration rate based on the environmental data and the transpiration demand model, the learning data creation device according to item 6. (Item 8) The plant physiological and ecological model includes the FvCB model, The environmental data includes photosynthetic photon flux density, ambient air temperature, ambient humidity, leaf surface boundary layer conductance for heat transport, and ambient carbon dioxide concentration, The individual leaf transpiration rate calculation unit calculates the individual leaf transpiration rate based on the environmental data and the FvCB model, the learning data creation device according to item 6. (Item 9) A learning unit that learns an estimation algorithm for estimating the transpiration rate of the plant based on the learning data created by the learning data creation device according to any one of items 1 to 8, A machine learning device comprising: (Item 10) The estimation algorithm is configured using an artificial neural network, the machine learning device according to item 9. (Item 11) An environmental data acquisition unit that acquires environmental data in a field, A plant image acquisition unit that acquires an image including leaves of a plant cultivated in the field, An individual leaf transpiration rate calculation unit that calculates an individual leaf transpiration rate based on the environmental data, An index calculation unit that calculates an index related to the leaf based on the image, A transpiration rate estimation unit that estimates the transpiration rate of the plant based on the individual leaf transpiration rate and the index related to the leaf according to the estimation algorithm learned by the machine learning device according to item 9 or 10, An estimation device comprising: (Item 12) A learning data creation method for creating learning data used for learning an estimation algorithm for estimating the transpiration rate of a plant, An individual leaf transpiration rate calculation step of calculating an individual leaf transpiration rate based on environmental data in a field, An index calculation step of calculating an index related to the leaf based on an image including the leaf of a plant cultivated in the field; A learning data creation step of creating learning data by associating the transpiration rate of the plant measured in the field with the individual leaf transpiration rate and the index related to the leaf; A learning data creation method including the above. (Item 13) A learning step of learning an estimation algorithm for estimating the transpiration rate of the plant based on the learning data created by the learning data creation method according to Item 12; A machine learning method including the above. (Item 14) An environmental data acquisition step of acquiring environmental data in the field; A plant image acquisition step of acquiring an image including the leaf of a plant cultivated in the field; An individual leaf transpiration rate calculation step of calculating an individual leaf transpiration rate based on the environmental data; An index calculation step of calculating an index related to the leaf based on the image; A transpiration rate estimation step of estimating the transpiration rate of the plant based on the individual leaf transpiration rate and the index related to the leaf according to the estimation algorithm learned by the machine learning method according to Item 13; An estimation method including the above. (Item 15) To a computer, A program for causing the computer to execute each step of the learning data creation method according to Item 12. (Item 16) To a computer, A program for causing the computer to execute each step of the machine learning method according to Item 13. (Item 17) To a computer, A program for causing the computer to execute each step of the estimation method according to Item 14.

Advantages of the Invention

[0011] According to the present invention, the transpiration rate of a plant can be quantitatively grasped.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Mode for Carrying Out the Invention

[0013] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and drawings, the same reference numerals denote the same or similar components, and thus duplicate descriptions of the same or similar components will be omitted. [First Embodiment]

[0014] In the present invention, the transpiration rate of individual leaves is calculated based on a plant physiological and ecological model and environmental data, and the transpiration rate of the plant is estimated using the calculated transpiration rate of individual leaves and an index related to the leaves. In the first embodiment, an Evaporative Demand (ED) model is used as the plant physiological and ecological model, and as environmental data, photosynthetic photon flux density (PPFD), the air temperature T a of the atmosphere, the humidity VPD a of the atmosphere, and the leaf surface boundary layer conductance g aH for heat transport are used. In the first embodiment, the CO2 concentration C a of the atmosphere is not used as environmental data. The transpiration demand model used in the first embodiment is a model that does not use the CO2 concentration C l of the atmosphere for calculating the transpiration rate Tr a of individual leaves. [Estimation Support System] [Overview of the System]

[0015] FIG. 1 is a diagram schematically showing a schematic configuration of a transpiration rate estimation support system according to the first embodiment of the present invention.

[0016] A transpiration rate estimation support system 100 (hereinafter, also simply referred to as the estimation support system 100) according to the first embodiment of the present invention includes a learning data creation device 1, a machine learning device 2, and an estimation device 3. The estimation support system 100 enables a producer in a small-scale farm 80 to quantitatively grasp the transpiration rate per plant of a plant 81 in the farm 80 operated by the producer without using an expensive measurement system that requires advanced measurement technology, such as a transparent open-type chamber surrounding the plant.

[0017] The learning data creation device 1, the machine learning device 2, and the estimation device 3 are directly or indirectly connected by wire or wirelessly in a manner capable of transmitting and receiving data via, for example, a network 10.

[0018] In this embodiment, the learning data creation device 1 and the machine learning device 2 are used by, for example, an organization or person (hereinafter also simply referred to as the administrator) that manages the test research field 90. The estimation device 3 is used by, for example, a producer (hereinafter also simply referred to as the producer) in a small-scale field 80. The administrator of the field 90 is, for example, a national or prefectural agricultural experiment station, a national public or private research institution or university, etc., and researchers in those organizations. The administrator has a measurement technique capable of measuring the data 41, 42, 43 described later using an open-type chamber installed in the field 90.

[0019] The plants 91, 81 cultivated in the fields 90, 80 will be described. The type of the plant 81 cultivated in the producer's field 80, which is the object to quantitatively grasp the transpiration rate per plant, is the same as the type of the plant 91 cultivated in the administrator's field 90. The plants cultivated in the fields 90, 80 are not limited to edible crops such as eggplants and leeks, and can be, for example, cut flowers. That is, the plants 91, 81 cultivated in the fields 90, 80 may be plants that transpire moisture from leaves.

[0020] The environmental data 41, the image 42 including the leaves of the plant 91, and the measurement data 43 of the transpiration rate per plant of the plant 91 used for creating the learning data 4 are acquired in the field 90 where a transparent open-type chamber surrounding the plant 91 is installed. The acquired data 41, 42, 43 are input to and recorded by the learning data creation device 1. The acquisition of the data 41, 42, 43 in the field 90 will be described later.

[0021] The learning data creation device 1 creates the learning data 4 used for learning the estimation algorithm 5 for estimating the transpiration rate per plant. The learning data 4 is created based on the data 41, 42, 43 measured in the test research field 90. The created learning data 4 is provided to the machine learning device 2. The learning data 4 and the estimation algorithm 5 are created for each plant type.

[0022] The machine learning device 2 learns the estimation algorithm 5 based on the learning data 4 created by the learning data creation device 1. The learned estimation algorithm 5 is provided to the estimation device 3.

[0023] In the producer's field 80, environmental data in the field 80 and an image including the leaves of the plant 81 are acquired using the sensing device 6 and the imaging device 7.

[0024] The estimation device 3 estimates the transpiration rate per plant of the plant 81 based on the individual leaf transpiration rate and an index related to the leaf according to the estimation algorithm 5 learned by the machine learning device 2. The individual leaf transpiration rate and the index related to the leaf are calculated based on the environmental data acquired in the producer's field 80 and the image including the leaves of the plant 81.

[0025] As a result, producers in small-scale fields 80 can quantitatively grasp the transpiration rate per plant of the plant 81 in the fields 80 operated by the producers without using expensive measurement systems that require advanced measurement techniques, such as transparent open chambers surrounding the plants.

[0026] In the experimental research field 90, the plant 91 is surrounded by a transparent open chamber 92 for measuring the transpiration rate. The chamber 92 is provided with an air inlet 93 and an air outlet 94, and outside air is ventilated into the chamber 92. The H2O concentration in the chamber 92 changes as water evaporates from the leaves of the plant 91 while the outside air passes through the chamber 92. Similarly, the CO2 concentration in the chamber 92 changes due to photosynthesis by the plant 91 while the outside air passes through the chamber 92. The gas analyzer 95 can measure the concentrations of water vapor (H2O) and CO2 for each of the air inlet 93 and the air outlet 94. The flow meter 96 provided at the air inlet 93 can measure the flow rate of the outside air flowing into the chamber 92.

[0027] The acquisition of data 41, 42, 43 in the field 90 will be described. In the present embodiment, as environmental data 41, photosynthetic photon flux density (PPFD) (unit: [μmol·m -2 ·s -1 ), the air temperature T a (unit: [°C]), the vapor pressure deficit (VPD) a (unit: [kPa]), and the leaf boundary layer conductance g aH for heat transport (unit: [mol·m -2 ·s -1 ) are used. In the present embodiment, the CO2 concentration C a (unit: [μmol·mol -1 ) of the atmosphere is not used as environmental data 41.

[0028] In the present embodiment, the photosynthetic photon flux density PPFD is measured using a PPFD sensor 97. The air temperature is measured using thermocouples 98a, 98b. The humidity is calculated from the concentration of water vapor measured by a gas analyzer 95 and the air temperature measured using thermocouples 98a, 98b. The leaf boundary layer conductance for heat transport is estimated from the wind speed of the atmosphere measured by, for example, a hot-wire anemometer 98c. In other embodiments, the concentration of CO2 is measured using a gas analyzer 95.

[0029] In the present embodiment, an image 42 including the leaves of the plant 91 cultivated in the field 90 is captured using an imaging device 99 (digital camera 99) attached to the ceiling surface of the chamber 92, and image data is acquired. The image 42 is a nadir view image looking down on the plant 91 from the ceiling surface toward the ground (soil), and is an image in which the leaves of the plant 91 are included in the imaging range. The color space of the image 42 captured by the imaging device 99 is in the RGB format.

[0030] In the present embodiment, the transpiration rate 43 per plant of the plant 91 is calculated from the product of the concentration difference of H2O between the intake port 93 and the exhaust port 94 and the flow rate of outside air flowing into the chamber 92.

[0031] The transpiration rate 43 calculated from the data measured using the chamber 92 is the transpiration rate for a plurality of strains of the plant 91 cultivated in the chamber 92, and such a transpiration rate is called the transpiration rate of the canopy. When the canopy is composed of a plurality of strains, the transpiration rate of the canopy is the sum of the transpiration rates per strain for each of the plurality of strains. In the field 90 where the chamber 92 is installed, the value of the transpiration rate of the canopy is the value obtained by integrating the values of the transpiration rates per strain for the plurality of strains of the plant 91 surrounded by the chamber 92. Therefore, the transpiration rate per strain of the plant 91 can be obtained by dividing the transpiration rate measured using the chamber 92 by the number of strains of the plant 91 surrounded by the chamber 92.

[0032] In agricultural cultivation, cases where the yield of crops, etc. are compared per unit land area are common. The transpiration rate per unit land area can be calculated by multiplying the transpiration rate per strain by the planting density (number of strains / m 2 ). <Hardware Configuration>

[0033] The learning data creation device 1, the machine learning device 2, and the estimation device 3 can be configured using, for example, a general-purpose computer, a tablet PC, a smartphone, or the like. All of these learning data creation device 1, machine learning device 2, and estimation device 3 can be configured using a general-purpose computer, or a part of them can be configured using a tablet PC or a smartphone.

[0034] For example, when using a smartphone, an integrated estimation device in which the estimation device 3, the imaging device 7, the display device 8, and the input device 9 described later are integrated can be configured. By communicably connecting a separate sensing device 6 to such an integrated estimation device using a smartphone by means of a wireless communication method such as Wi-Fi (registered trademark) or Bluetooth (registered trademark), a user terminal that can be easily used by producers at the agricultural site can be provided.

[0035] Each of the training data creation device 1, the machine learning device 2, and the estimation device 3 includes, as hardware configuration, a processor (not shown) such as a CPU that processes data, a main storage device (not shown) that the processor uses as a working area for data processing, and auxiliary storage devices 19, 29, and 39 that are used for temporarily storing data. Data 41, 42, and 43, training data 4, an estimation algorithm 5, a training data creation program, a machine learning program, an estimation program, and the like are appropriately stored in the auxiliary storage devices 19, 29, and 39, respectively.

[0036] A sensing device 6 and an imaging device 7 are connected to the estimation device 3 as hardware components. The sensing device 6 and the imaging device 7 are used by a producer, who is a user of the estimation device 3, to acquire environmental data in a farm field 80 and an image including leaves of a plant 81.

[0037] In this embodiment, the sensing device 6 measures the photosynthetic photon flux density PPFD, the atmospheric temperature, the atmospheric humidity, and the leaf surface boundary layer conductance to heat transport as environmental data in the field 80. The sensing device 6 is equipped with various sensors for measuring each of these environmental data. For example, a known PPFD sensor can be used as the sensor for measuring the photosynthetic photon flux density PPFD. A known temperature and humidity sensor can be used as the sensor for measuring the temperature and humidity. For example, a hot wire anemometer can be used as the sensor for estimating the leaf surface boundary layer conductance to heat transport. Leaf surface boundary layer conductance g aH can be estimated from the wind speed measured by an anemometer. Assuming that the wind is blowing in laminar and forced convection on the upper and lower surfaces of the leaf, g aH =6.62×10 -3 (u / d), where u is the wind speed (unit: m s -1) where d is the characteristic length of the leaf (unit: [m]). In other embodiments, the sensing device 6 can further measure the concentration of CO2 in the atmosphere. A known CO2 sensor can be used as the sensor for measuring the concentration of CO2. All of these sensors can be realized by portable devices, and the sensing device 6 can be easily used.

[0038] The imaging device 7 captures an image including the leaves of the plant 81 cultivated in the field 80 and acquires image data. Similar to the image 42 acquired in the field 90, in this embodiment, the image captured by the imaging device 7 is a directly downward view image of the plant 81, and the image includes almost all the leaves of one plant 81 within the imaging range. A digital camera 7 can be used as the imaging device 7.

[0039] As shown in FIG. 6 to be described later, in this embodiment, the estimation device 3 includes a display device 8. For example, a liquid crystal monitor can be used as the display device 8 to display information to the producer who is the user of the estimation device 3. Further, an input device 9 can be connected to the estimation device 3 as an optional configuration. For example, a keyboard, a touch panel, a mouse, etc. can be used as the input device 9 to receive input operations from the user. The estimation device 3 can also acquire, for example, the values of the environmental data measured by the user using the sensing device 6 via the input device 9. A touch panel in which the display device 8 and the input device 9 are integrated may be connected to the estimation device 3. [Learning Data Creation Device]

[0040] FIG. 2 is a block diagram for explaining the functions of the learning data creation device and the machine learning device according to the first embodiment of the present invention.

[0041] The learning data creation device 1 includes, as functional blocks, a single-leaf transpiration rate calculation unit 11, an index calculation unit 12, and a learning data creation unit 13. The index calculation unit 12 includes, as functional blocks, a leaf area extraction unit 121, a leaf area index calculation unit 122, and a light reception efficiency index calculation unit 123. These functional blocks can be implemented as hardware using an integrated circuit or the like. Alternatively, these functional blocks can also be implemented as software by the processor of the learning data creation device 1 reading and executing a learning data creation program in the main storage device of the learning data creation device 1.

[0042] The single-leaf transpiration rate calculation unit 11 calculates the single-leaf transpiration rate based on the environmental data 41 in the field 90. In the present embodiment, the single-leaf transpiration rate calculation unit 11 calculates the single-leaf transpiration rate Tr l based on the environmental data 41 and a plant physio-ecological model.

[0043] FIG. 3 is a diagram for explaining the evapotranspiration demand (ED) model of the plant physio-ecological model introduced in the first embodiment of the present invention. (A) is a graph showing the changes during the day of the single-leaf transpiration rate Tr l , the evapotranspiration demand degree ED, and the impact coefficient imp used in the evapotranspiration demand model. (B) is a graph showing the relationship between the impact coefficient imp and the photosynthetic photon flux density PPFD.

[0044] In the present embodiment, the plant physio-ecological model includes an evapotranspiration demand (ED) model, and calculates the single-leaf transpiration rate Tr l based on the evapotranspiration demand model. As represented by Equations 1 to 4 described later, in the evapotranspiration demand model, the single-leaf transpiration rate Tr lIt is expressed using the product of the transpiration demand degree ED and the impact coefficient imp. As shown in Fig. 3(B), the impact coefficient imp is approximated as a non-rectangular hyperbola of the photosynthetic photon flux density PPFD. As the sensor for measuring the photosynthetic photon flux density PPFD, a known PPFD sensor can be used, and the photosynthetic photon flux density PPFD can be obtained as an actual measurement value. Thereby, in the plant physiological and ecological model, instead of the model parameters that are difficult to measure, parameters that can be actually measured are used to calculate the transpiration rate Tr l of each leaf.

[0045] Note that when performing learning of the estimation algorithm, by setting learning data with high explicability for the output layer 53 in the input layer 51, even if the amount of the learning data 4 is smaller than before, estimation with high accuracy becomes possible. That is, the more the logical causal relationship between the data set in the input layer 51 and the data set in the output layer 53 is improved, the learning of the estimation algorithm 5 can be performed with a smaller amount of learning data.

[0046] The transpiration demand degree ED defines the transpiration demand from the environment to each leaf and is expressed by the following formula 1. According to the transpiration demand degree ED, the net radiation flux density on the leaf surface, which is difficult to actually measure, is expressed by environmental factors that can be measured (for example, short-wave radiation flux density R s , air temperature T A , air saturation deficit VPD a etc.). More specifically, the transpiration demand degree ED represents the evaporation rate from a free water surface when it is assumed that the transpiration regulation function by stomata and cuticle layer does not exist on the leaf surface.

Equation

[0047] The impact coefficient imp is a limiting coefficient (0 ≤ imp ≤ 1) representing the transpiration regulation function of the leaf and is expressed by the following formulas 2 to 3.

Equation

[0048] Using the evapotranspiration requirement degree ED and the impact coefficient imp, the transpiration rate Tr of an individual leaf l is represented by the following Equation 4. As represented by Equation 4, the impact coefficient imp is a coefficient that determines the transpiration flow caused by the evapotranspiration requirement degree ED from the environmental side for the plant. The impact of the evapotranspiration requirement degree ED from the environmental side on the plant water balance becomes higher as the impact coefficient imp becomes larger.

Equation

[0049] The variables represented in Equations 1 to 4 will be described.

[0050] Δ is the gradient of the humidity T on the saturation water vapor concentration curve, and the humidity T E is the equilibrium temperature between the leaf and the environment. α is the absorption of the leaf with respect to shortwave radiation, and is approximately 0.5 for sunlight. R E is the shortwave radiation flux density. σ is the Stefan-Boltzmann constant. ε S is the emissivity of the environment, and ε A is the emissivity of the leaf. T L is the air temperature. C A is the volumetric heat capacity of air. γ is the psychrometric constant of air. n is the number of surfaces considering transpiration, which is a constant (n = 2 for both-sided stomata and n = 1 for one-sided stomata), and is selected according to the plant species. λ is the latent heat of vaporization of water. p ρ is the volumetric heat capacity of air. γ is the psychrometric constant of air. n is the number of surfaces considering transpiration, which is a constant (n = 2 for both-sided stomata and n = 1 for one-sided stomata), and is selected according to the plant species. λ is the latent heat of vaporization of water.

[0051] g e ,g s ,g aw is the leaf conductance (unit [mol·m -2 ·s -1 ). g e is the sensible heat conductance at the leaf surface. g s is the stomatal conductance. g aw is the leaf surface boundary layer conductance for water vapor transport.

[0052] The shortwave radiation flux density R S , the air temperature T A, the vapor pressure deficit VPD a , and the leaf boundary layer conductance g for water vapor transport aw are obtained, other physical quantities can be obtained, and it is possible to evaluate the evapotranspiration demand ED. Among the variables expressed in Equations 1 to 4, the variables whose values are determined in advance by other measurements or literature values are α, σ, ε A , ε L , γ, C p ρ, λ.

[0053] In addition, in FIG. 3(B), the plotted values of the impact coefficient imp are values evaluated according to the solar radiation intensity (i.e., PPFD). The solid-line curve shows that the impact coefficient imp can be approximated as a function of the photosynthetic photon flux density PPFD using the following Equations 5 to 6. [Number]

[0054] Fitting is performed by the non-linear least squares method. The variables imp max , φ, and θ in Equations 5 to 6 are fitting parameters. In the example shown in FIG. 3(B), imp max = 0.92, φ = 800, θ = -86400. As shown in FIG. 3(B), the impact coefficient imp can be approximated as a non-rectangular hyperbola of the photosynthetic photon flux density PPFD.

[0055] Referring to FIG. 2 again. The index calculation unit 12 calculates an index related to the leaf based on the image 42 including the leaf of the plant 91 cultivated in the field 90. In the present embodiment, as an index related to the leaf, the leaf area index (Leaf Area Index; LAI) and the light reception efficiency index of the leaf are calculated. Prior to the calculation of these indices, the leaf region extraction unit 121 extracts the region of the leaf.

[0056] The leaf area extraction unit 121 extracts the leaf area in the image 42 including the leaves of the plant 91 by image analysis using artificial intelligence. In the present embodiment, the leaf area extraction unit 121 extracts the leaf area in the image 42 based on a known image recognition method using deep learning, which is an example of artificial intelligence. The leaf area extraction unit 121 replaces the area determined to be an area other than the leaf with black. As will be described later, the value of the relative luminance L of the black area is zero.

[0057] FIG. 4 is a diagram showing an example in which the leaf area is extracted by image analysis using artificial intelligence. (A) is an image before the leaf area is extracted, and (B) is an image after the leaf area is extracted by artificial intelligence.

[0058] In (A) and (B), the area indicated by reference numeral 101 is the leaf area of the plant. The area indicated by reference numeral 102 in (A) is a vinyl sheet covering the soil for cultivating the plant. For example, the leaf area extraction unit 121 discriminates, by image analysis using artificial intelligence, an area such as this vinyl sheet indicated by reference numeral 102 in (A) as an area other than the leaf. An area discriminated to be an area other than the leaf, such as the area of this vinyl sheet, is displayed in black as indicated by reference numeral 103 in (B).

[0059] Referring to FIG. 2 again. The leaf area index calculation unit 122 calculates the leaf area index LAI using the ratio occupied by the area other than the leaf in the image 42 obtained by binarizing the image 42 including the leaves of the plant 91. Exemplarily, binarization of an image means blackening the image. In the present embodiment, the image 42 including the leaves of the plant 91 is a directly downward view image looking down on the plant 91 from the ceiling surface toward the ground. By binarizing (i.e., blackening) such a directly downward view image, the image 42 including the leaves is divided into the leaf area a L and the area other than the leaf a NL When divided into two, the ratio P0 occupied by the area other than the leaf is P0 = a NL / (a NL + a L ). Using P0, the leaf area index LAI can be calculated based on the following formula 7.

Number

[0060] The calculation of the leaf area index LAI will be described in detail. Consider a vertical view image obtained by looking down from the ceiling surface towards the ground at a plant stock. In such a vertical view image, consider dividing a stock having a leaf area index of L into a sufficiently large number N of layers extending vertically downward from the surface to the bottom surface of the stock. Assume that a part of the leaves of the stock is included in each of the divided layers. At this time, the following four matters are assumed.

[0061] Assumption 1: The leaves are uniformly distributed azimuthally and randomly in space. Assumption 2: The leaf area index of each layer is equal. That is, L / N = ΔL. Assumption 3: The light rays are irradiated vertically downward, and the probability that the light rays contact the leaves multiple times in each layer is extremely small and zero compared to the probability that the light rays contact the leaves only once. Assumption 4: The probability that the light rays contact the leaves in each layer is equal in all layers based on Assumption 1. That probability is equal to the value obtained by horizontally projecting the leaf area index of each layer, and when the coefficient is G, it can be expressed as GΔL, and the probability that the light rays do not contact the leaves can be expressed as (1 - GΔL).

[0062] When the light rays pass through N layers from the surface to the bottom surface of the stock, the probability P0 that the light rays do not contact the leaves in all layers is described as follows in Equation 8 using the combination symbol C.

Number

[0063] Here, when N is made infinitely large, Equation 8 can be transformed as follows in Equation 9.

Number

[0064] Equation 9 shows that the probability P0 is related to the Poisson distribution of the leaf area index magnitude L. Here, assuming that the leaf inclination angles are spherically distributed, the coefficient G can be approximated as 0.5. By transforming Equation 9 with this approximation, the following Equation 10 can be obtained.

Number

[0065] From the ratio (probability P0) occupied by the area other than the leaves in the directly downward view image of the plant stock, the leaf area index LAI can be calculated by Equation 10.

[0066] The light reception efficiency index calculation unit 123 calculates the light reception efficiency index by integrating the relative luminance of the leaf area for the image 42 including the leaves of the plant 91.

[0067] The relative luminance is a value normalized with respect to the reference white. In an RGB format image, if the RGB values of a certain pixel are represented as R, G, and B respectively, the relative luminance L of that pixel can be calculated, for example, by the following Equation 11.

[0068] Relative luminance L = 0.2126×R + 0.7152×G + 0.0722×B (Equation 11)

[0069] The light reception efficiency index calculation unit 123 converts the RGB format image 42 into a luminance image, calculates the relative luminance for each pixel included in the leaf area in the converted luminance image, and integrates those relative luminances calculated for all pixels in the image 42. The integrated value of the relative luminance L calculated for the image 42 divided by the value of the photosynthetic photon flux density PPFD measured directly above the plant 91 is the light reception efficiency index. The light reception efficiency index can be calculated by the following Equation 12.

Number

[0070] In this embodiment, in the RGB-formatted image 42, regions other than the leaf regions have already been replaced with black by the leaf region extraction unit 121. The RGB values of black are R = 0, G = 0, and B = 0. That is, for the black regions, the value of the relative luminance L becomes zero, and for the regions other than the leaves that have been replaced with black in the RGB-formatted image 42, they do not contribute to the integrated value of the relative luminance L.

[0071] The learning data creation unit 13 creates learning data 4 by associating the transpiration rate per plant 43 of the plant 91 measured in the field 90 with the transpiration rate per individual leaf and the index related to the leaf. The learning data 4 is a data set used for the learning of the estimation algorithm 5, and is data in which the data set in the input layer 51 and the data set in the output layer 53 are set. The created learning data 4 is stored, for example, in the auxiliary storage device 19.

[0072] The created learning data 4 is transmitted from the learning data creation device 1 to the machine learning device 2 and stored in the auxiliary storage device 29 of the machine learning device 2. The machine learning device 2 performs machine learning of the estimation algorithm 5 using the received learning data 4. [Machine learning device]

[0073] Referring to FIG. 2 again. The machine learning device 2 includes a learning unit 21 as a functional block. The learning unit 21 can be implemented as hardware using an integrated circuit or the like. Alternatively, the learning unit 21 can also be implemented as software when the processor of the machine learning device 2 reads the machine learning program into the main storage device of the machine learning device 2 and executes it.

[0074] The learning unit 21 learns an estimation algorithm 5 for estimating the transpiration rate per plant of the plant based on the learning data 4 created by the learning data creation device 1.

[0075] FIG. 5 is a schematic diagram for explaining an artificial neural network used in the estimation algorithm according to the first embodiment of the present invention.

[0076] In this embodiment, the estimation algorithm 5 is configured using an Artificial Neural Network (ANN). The estimation algorithm 5 includes an input layer 51, an intermediate layer 52, and an output layer 53 as the layers that make up the neural network. The learning data 4 is composed of the transpiration rate per plant 43 of the plant 91, the individual leaf transpiration rate, and the index related to the leaf, each of which is associated with one another. The learning unit 21 sets the individual leaf transpiration rate and the index related to the leaf in the input layer 51, sets the transpiration rate per plant 43 in the output layer 53, and learns the estimation algorithm 5. The learned estimation algorithm 5 is stored in the auxiliary storage device 29.

[0077] When learning the estimation algorithm, by setting learning data with high explicability for the output layer 53 in the input layer 51, even if the amount of learning data 4 is smaller than before, estimation with high accuracy becomes possible. That is, the more the logical causal relationship between the data set in the input layer 51 and the data set in the output layer 53 improves, the more the learning of the estimation algorithm 5 can be performed with a smaller amount of learning data.

[0078] In this embodiment, a Feedforward Neural Network is used for the estimation algorithm 5. The neural network of the estimation algorithm 5 includes a plurality of layers in the intermediate layer 52. As the learning algorithm of the neural network, the back propagation method is used to adjust the weight parameters (synaptic weights) in the estimation algorithm 5.

[0079] The learned estimation algorithm 5 is transmitted from the machine learning device 2 to the estimation device 3 and stored in the auxiliary storage device 39 of the estimation device 3. The estimation device 3 estimates the transpiration rate per plant of the plant 81 in the producer's field 80 using the received learned estimation algorithm 5. [Estimation Device]

[0080] FIG. 6 is a block diagram for explaining the functions of the estimation device according to the first embodiment of the present invention.

[0081] The estimation device 3 includes, as functional blocks, an environmental data acquisition unit 31, a plant image acquisition unit 32, an individual leaf transpiration rate calculation unit 33, an index calculation unit 34, and a transpiration rate estimation unit 35. The index calculation unit 34 includes, as functional blocks, a leaf area extraction unit 341, a leaf area index calculation unit 342, and a light reception efficiency index calculation unit 343. These functional blocks can be implemented as hardware using an integrated circuit or the like. Alternatively, these functional blocks can also be implemented as software by a processor of the estimation device 3 reading an estimation program into the main storage device of the estimation device 3 and executing it.

[0082] A sensing device 6 and an imaging device 7 are connected to the estimation device 3. A user of the estimation device 3 uses the sensing device 6 and the imaging device 7 to acquire environmental data in the field 80 and an image including the leaves of the plant 81.

[0083] The environmental data acquisition unit 31 acquires environmental data in the field 80 from the sensing device 6. The types of environmental data to be acquired are the same as the environmental data 41 used for creating the learning data 4 in the learning data creation device 1. In the present embodiment, using the sensing device 6 and the imaging device 7, as environmental data in the field 80, the photosynthetic photon flux density PPFD, the air temperature T of the atmosphere a and the vapor pressure deficit VPD of the atmosphere a and the leaf surface boundary layer conductance g for heat transport aH are acquired.

[0084] The plant image acquisition unit 32 acquires, from the imaging device 7, an image including the leaves of the plant 81 being cultivated in the field 80. In the present embodiment, the method of acquiring an image including the leaves of the plant 81 is the same as the method of acquiring an image including the leaves of the plant 91 in the field 90. That is, the image including the leaves of the plant 81 to be imaged is imaged using the imaging device 7, and image data is acquired. The acquired image is a directly downward view image looking down on the plant 81 from above toward the ground, and is an image in which substantially all the leaves of one plant 81 are included in the imaging range. The color space of the image imaged by the imaging device 7 is in the RGB format.

[0085] The individual leaf transpiration rate calculation unit 33 has the same function as the individual leaf transpiration rate calculation unit 11 of the learning data creation device 1 shown in FIG. 2. The individual leaf transpiration rate calculation unit 33 calculates the individual leaf transpiration rate based on the environmental data in the field 80 acquired by the environmental data acquisition unit 31.

[0086] The index calculation unit 34 has the same function as the index calculation unit 12 of the learning data creation device 1 shown in FIG. 2. That is, the leaf area extraction unit 341, the leaf area index calculation unit 342, and the light reception efficiency index calculation unit 343 provided in the index calculation unit 34 have the same functions as the leaf area extraction unit 121, the leaf area index calculation unit 122, and the light reception efficiency index calculation unit 123 provided in the index calculation unit 12.

[0087] The index calculation unit 34 calculates an index related to the leaves based on the image including the leaves of the plant 81 being cultivated in the field 80 acquired by the plant image acquisition unit 32. The leaf area extraction unit 341 extracts the leaf area in the image including the leaves of the plant 81 by image analysis using artificial intelligence. The leaf area index calculation unit 342 calculates the leaf area index LAI using the ratio occupied by the area other than the leaves in the image obtained by binarizing the image including the leaves of the plant 81. The light reception efficiency index calculation unit 343 calculates the light reception efficiency index by integrating the relative luminance of the leaf area for the image including the leaves of the plant 81.

[0088] The transpiration rate estimation unit 35 estimates the transpiration rate per plant of the plant 81 based on the transpiration rate of each leaf and the index related to the leaf according to the estimation algorithm 5 learned by the machine learning device 2.

[0089] The auxiliary storage device 39 stores the learned estimation algorithm 5. The transpiration rate estimation unit 35 inputs the transpiration rate of each leaf calculated by the transpiration rate calculation unit 33 of each leaf and the index related to the leaf calculated by the index calculation unit 34 into the input layer 51 of the learned estimation algorithm 5, whereby an estimated value of the transpiration rate per plant of the plant 81 is output from the output layer 53. In the present embodiment, the obtained estimated value output from the output layer 53 is displayed on the display device 8 and presented to the producer who is the user of the estimation device 3.

[0090] Thereby, producers in small-scale farms 80 can quantitatively grasp the transpiration rate per plant of the plant 81 in the farms 80 operated by the producers without using an expensive measurement system that requires advanced measurement technology, such as a transparent open-type chamber surrounding the plants. [Machine learning procedure]

[0091] FIG. 7 and FIG. 8 are flowcharts for explaining the procedure of machine learning of the estimation algorithm using the machine learning device according to the first embodiment of the present invention.

[0092] The machine learning procedure is performed, for example, by a researcher at an agricultural experiment station who is the administrator of a test and research farm 90, using the learning data creation device 1 and the machine learning device 2.

[0093] The machine learning procedure includes steps S1 to S7 of creating learning data 4 used for learning the estimation algorithm 5 and step S8 of learning the estimation algorithm 5 based on the created learning data 4. The procedures of steps S1 to S7 are performed using the learning data creation device 1, and the procedure of step S8 is performed using the machine learning device 2.

[0094] In step S1, environmental data 41 in the field 90 is acquired. In step S2, an image 42 including the leaves of the plant 91 cultivated in the field 90 is acquired. In step S3, measurement data 43 of the transpiration rate per plant of the plant 91 in the field 90 is acquired. Exemplarily, the environmental data 41 and the measurement data 43 of the transpiration rate are measured every 30 minutes, and a total of 144 sets for 3 days are acquired. The image 42 including the leaves is captured every day, and a total of 3 days' worth is acquired. Note that in the image 42 including the leaves, the change in the value of the leaf area index LAI calculated by Equation 10 is negligibly small within a day. Therefore, for the set of the environmental data 41 and the measurement data 43 of the transpiration rate measured on the same day, the image 42 captured on the same day as the day when these data 41 and 43 were measured can be reused.

[0095] As shown in FIG. 1, the environmental data 41 is acquired, for example, by a researcher at an agricultural experimental station using an open chamber 92 installed in the field 90. The image 42 including the leaves of the plant 91 is acquired using an imaging device 99 attached to the ceiling surface of the chamber 92. The transpiration rate per plant 43 is calculated from the product of the concentration difference of H2O between the intake port 93 and the exhaust port 94 and the flow rate of the outside air flowing into the chamber 92. The acquired environmental data 41, the image data 42 including the leaves of the plant 91, and the measurement data 43 of the transpiration rate 43 per plant are stored in the auxiliary storage device 19 of the learning data creation device 1.

[0096] In step S4 (individual leaf transpiration rate calculation step), the individual leaf transpiration rate calculation unit 11 calculates the individual leaf transpiration rate based on the acquired environmental data 41.

[0097] In step S5 (index calculation step), the index calculation unit 12 calculates an index related to the leaf based on the acquired image data 42 including the leaves of the plant 91. In step S5, the procedures of step S5a to step S5c described below are performed.

[0098] In step S5a (leaf area extraction step), the leaf area extraction unit 121 extracts the leaf area in the image 42 including the leaves of the plant 91 by image analysis using artificial intelligence.

[0099] In step S5b (leaf area index calculation step), the leaf area index calculation unit 122 calculates the leaf area index using the ratio occupied by the area other than the leaves in the image 42, which is obtained by binarizing the image 42 including the leaves of the plant 91.

[0100] In step S5c (light reception efficiency index calculation step), the light reception efficiency index calculation unit 123 calculates the light reception efficiency index by integrating the relative luminance of the leaf area for the image 42 including the leaves of the plant 91.

[0101] In step S6 (learning data creation step), the learning data creation unit 13 creates the learning data 4 by associating the transpiration rate per plant 43 of the plant 91 measured in the field 90 with the individual leaf transpiration rate calculated in step S4 and the index related to the leaves calculated in step S5. The created learning data 4 is stored in the auxiliary storage device 19.

[0102] Also, the learning data 4 is transmitted from the learning data creation device 1 to the machine learning device 2 and stored in the auxiliary storage device 29 of the machine learning device 2. The machine learning device 2 performs machine learning of the estimation algorithm 5 using the received learning data 4.

[0103] In step S7, for example, the learning data creation device 1 itself determines whether the number of the learning data 4 is sufficient. If the number of the learning data 4 is not sufficient, the procedures of steps S1 to S6 are repeated. If the number of the learning data 4 is sufficient, the procedure of step S8 is performed. Exemplarily, the number of the learning data 4 required to perform the procedure of step S8 is about 144 sets.

[0104] In step S8 (learning step), the learning unit 21 learns an estimation algorithm 5 for estimating the transpiration rate per plant based on the learning data 4 created in the learning data creation device 1. The learned estimation algorithm 5 is stored in the auxiliary storage device 29.

[0105] Also, the learned estimation algorithm 5 is transmitted from the machine learning device 2 to the estimation device 3 and stored in the auxiliary storage device 39 of the estimation device 3. The estimation device 3 estimates the transpiration rate per plant of the plant 81 in the producer's field 80 using the received learned estimation algorithm 5. [Estimation procedure]

[0106] FIG. 9 and FIG. 10 are flowcharts for explaining the procedure for estimating the transpiration rate using the estimation device and the learned estimation algorithm according to the first embodiment of the present invention.

[0107] The estimation procedure is performed, for example, by a producer in a small-scale field 80 using the estimation device 3. The estimation procedure includes the following steps S11 to S15.

[0108] In step S11 (environmental data acquisition step), environmental data in the field 80 is acquired. In step S12 (plant image acquisition step), an image including the leaves of the plant 81 cultivated in the field 80 is acquired.

[0109] As shown in FIG. 1, the environmental data and the image including the leaves of the plant 81 are acquired, for example, by a producer in a farm cultivating the plant 81 using the sensing device 6 and the imaging device 7 in the field 80.

[0110] In step S13 (individual leaf transpiration rate calculation step), the individual leaf transpiration rate calculation unit 33 calculates the individual leaf transpiration rate based on the acquired environmental data.

[0111] In step S14 (index calculation step), the index calculation unit 34 calculates an index related to the leaves based on the acquired image including the leaves of the plant 81. In step S14, the procedures of steps S14a to S14c described below are performed.

[0112] In step S14a (leaf area extraction step), the leaf area extraction unit 341 extracts the leaf area in the image including the leaves of the plant 81 by image analysis using artificial intelligence.

[0113] In step S14b (leaf area index calculation step), the leaf area index calculation unit 342 calculates the leaf area index using the ratio occupied by the area other than the leaves in the image obtained by binarizing the image including the leaves of the plant 81.

[0114] In step S14c (light reception efficiency index calculation step), the light reception efficiency index calculation unit 343 calculates the light reception efficiency index by integrating the relative luminance of the leaf area for the image including the leaves of the plant 81.

[0115] In step S15 (transpiration rate estimation step), the transpiration rate estimation unit 35 estimates the transpiration rate per plant of the plant 81 based on the transpiration rate of each leaf calculated in step S13 and the index related to the leaves calculated in step S14 according to the estimation algorithm 5 learned by the machine learning method. The obtained estimated value is displayed on, for example, the display device 8 and presented to the producer who is the user of the estimation device 3.

[0116] Thereby, producers in the small-scale farm 80 can quantitatively grasp the transpiration rate per plant of the plant 81 in the farm 80 operated by the producer without using an expensive measurement system that requires advanced measurement technology, such as a transparent open chamber surrounding the plant. [Effect]

[0117] As described above, according to the learning data creation device, machine learning device, and estimation device according to the first embodiment of the present invention, as well as the learning data creation method, machine learning method, and estimation method, the transpiration rate of plants can be quantitatively grasped.

[0118] As a result, producers in small-scale farms 80 can quantitatively grasp the transpiration rate per plant 81 in the farms 80 operated by the producers without using expensive measurement systems that require advanced measurement techniques, such as transparent open chambers surrounding the plants.

[0119] When producers can quantitatively grasp the transpiration rate per plant 81 in the farm 80, it becomes possible to grasp the growth status and yield of the plants 81 in the farming field without relying on the producers' experience and intuition. As a result, for example, it becomes possible to appropriately determine the timing of fertilization, irrigation, supplementary lighting, and CO2 application in the farming field, promoting the efficiency of agriculture.

[0120] For the estimation device 3 used by producers for estimation in the farming field, devices such as tablet PCs and smartphones can be used. Furthermore, by communicably connecting the handheld sensing device 6 to such an estimation device 3, producers can easily estimate the transpiration rate per plant.

[0121] In addition, by daily performing such estimation of the transpiration rate in the farming field, producers can also grasp information on the transpiration rate per plant, which is a basic raw material for plant production, in a time series. As a result, for example, it becomes possible to more appropriately determine the timing of fertilization and irrigation described above in the farming field, further promoting the efficiency of agriculture. [Second Embodiment]

[0122] Unless otherwise specified, the configurations of the learning data creation device, machine learning device, and estimation device according to the second embodiment described below are the same as those of the learning data creation device, machine learning device, and estimation device according to the first embodiment, and thus duplicate explanations are omitted.

[0123] FIG. 11 is a schematic diagram for explaining an artificial neural network used in an estimation algorithm according to a second embodiment of the present invention.

[0124] In the second embodiment, the FvCB (Farquhar-von Caemmerer-Berry) model is used as a plant physiological and ecological model, and as environmental data, photosynthetic photon flux density (PPFD), atmospheric temperature T a , atmospheric humidity VPD a , leaf surface boundary layer conductance g for heat transport aH , and atmospheric CO2 concentration C a are used. In the second embodiment, atmospheric CO2 concentration C a is used as environmental data. The FvCB model used in the second embodiment is a model that uses atmospheric CO2 concentration C l to calculate the transpiration rate Tr a per individual leaf. The second embodiment is different from the first embodiment in that the FvCB model is used as a plant physiological and ecological model and the environmental data includes atmospheric CO2 concentration C a . The concentration of CO2 in the field 90 is measured using the gas analyzer 95, and the CO2 concentration in the field 80 is acquired using the sensing device 6.

[0125] According to the second embodiment of the present invention, at the time of creating learning data, the individual leaf transpiration rate calculation unit 11 calculates the individual leaf transpiration rate Tr l based on the environmental data 41 in the field 90 and the FvCB model. At the time of estimation, the individual leaf transpiration rate calculation unit 33 calculates the transpiration rate per plant of the plant 81 based on the environmental data in the field 80 and the FvCB model.

[0126] Hereinafter, the FvCB model of the plant physiological and ecological model introduced in the second embodiment will be described in detail.

[0127] Figure 12 is a schematic diagram for explaining the FvCB model of the plant physiological and ecological model introduced in the second embodiment of the present invention. (A) shows a chloroplast, and (B) shows a cross-sectional view of a leaf.

[0128] In this embodiment, the FvCB model includes a biochemical model of photosynthesis, a transport equation, a stomatal conductance model, and a heat balance model. The transport equation includes a transport equation for CO2 and a transport equation for H2O. The biochemical model of photosynthesis is related to the action in the chloroplast shown in (A). The transport equation, the stomatal conductance model, and the heat balance model are related to the action in the leaf shown in (B). In this embodiment, by substituting the value of the environmental data 41 into each of the following mathematical formulas representing these four models and solving the simultaneous equations, the transpiration rate Tr of an individual leaf is calculated from Equation 19. l is calculated.

[0129] The biochemical model of photosynthesis is represented by the following Equations 13 to 16.

Equation

[0130] In the biochemical model of photosynthesis, the photosynthesis rate is divided into two photosynthesis rates at two rate-limiting stages: the photosynthesis rate A at the Rubisco-limited stage where the rate is limited by the CO2 concentration and the photosynthesis rate A at the RuBP-limited stage mainly rate-limited by the light intensity. Among these two photosynthesis rates A L,c and A L,j , the lower one is taken as the photosynthesis rate A of an individual leaf. L,c ,A L,j of the two, and the lower one is taken as the photosynthesis rate A of an individual leaf. L is used.

[0131] Note that the mesophyll conductance g shown in (B) of Figure 12 is difficult to measure. Therefore, in the biochemical model of photosynthesis introduced in this embodiment, the CO2 concentration C in the chloroplast m and the CO2 concentration C between the intracellular spaces in the leaf c and the CO2 concentration C between the intracellular spaces in the leaf iAssuming there is almost no difference between them, C c = C i is set as such.

[0132] The transport equation for CO2 is represented by the following equations (17) and (18).

Equation

[0133] The transport equation for H2O is represented by the following equation (19).

Equation

[0134] The stomatal conductance model is represented by the following equation (20).

Equation

[0135] The heat balance model is represented by the following equation (21).

Equation

[0136] The variables represented in equations (13) to (21) and FIG. 12 will be described.

[0137] PPFD, C a , T a , and VPD a are the photosynthetic photon flux density PPFD, the CO2 concentration in the atmosphere, the atmospheric temperature, and the atmospheric humidity, respectively, and are acquired as environmental data at the fields 90, 80.

[0138] A L , A L,c , A L,j means the photosynthesis rate (unit: [μmol·m -2 ·s -1 ). A L is the photosynthesis rate of an individual leaf. A L,cis the photosynthesis rate at the Rubisco-limited stage. A L,j is the photosynthesis rate at the RuBP-limited stage.

[0139] V cmax is the maximum rate of carboxylation (unit [μmol·m -2 ·s -1 ) and is a value determined for each plant species. R d is the day respiration rate (unit [μmol·m -2 ·s -1 ). Γ * is the CO2 compensation point of CO2 assimilation in the absence of the day respiration rate R d (unit [μmol·m -2 ·s -1 ).

[0140] C i ,C c ,C s means the CO2 concentration in the leaf (unit [μmol·mol -1 ). C i is the CO2 concentration between leaf cells. C c is the CO2 concentration in the chloroplast. C s is the CO2 concentration on the leaf surface. O is the O2 concentration between leaf cells (unit [mol·mol -1 ).

[0141] K c ,K O is the Michaelis-Menten constant (unit [μmol·mol -1 ). K c is the Michaelis-Menten constant for carboxylation. K O is the Michaelis-Menten constant for carboxylation oxygenation.

[0142] J is the electron transport rate (unit [μmol·m -2 ·s -1 ) and is a value determined for each plant species. J maxis the maximum value of the electron transfer rate J. θ is the convexity of the J-PPFD curve. φ is the initial slope of the J-PPFD curve.

[0143] g s ,g a ,g m represents the leaf conductance (unit [mol·m -2 ·s -1 ). g s is the stomatal conductance. g a is the leaf boundary layer conductance. g m is the mesophyll conductance. g aw is the leaf boundary layer conductance for water vapor transport, and g sw is the stomatal conductance for water vapor transport. g0, g1 are the fitting parameters of the stomatal conductance g s . VPD L is the saturation deficit between the leaf surface and the atmosphere (unit [Pa]). Pa is the atmospheric pressure.

[0144] ρ a is the air density (1.204 kg·m -3 ). C p is the specific heat of air at constant pressure (1010 J·kg -1 ·K -1 ).

[0145] T L ,γ,R ni ,s,g HR ,g LW are variables describing the heat balance model. T L is the leaf temperature (unit [K]). γ is the psychrometric constant (unit [Pa·K -1 ). R ni is the isothermal pure radiation (unit [W·m -2 ·s -1 ). s is the slope of the saturated water vapor pressure with respect to temperature (unit [Pa·K -1 ). g HR is the total conductance of radiation and sensible heat transport in the leaf boundary layer (unit [m·s -1 ). gLW is the combined conductance of water molecule transport in the leaf surface boundary layer and stomata (unit: [m·s -1 ).

[0146] Among the variables expressed in Equations 13 to 21, the variables whose values are determined in advance by other measurements and their values are shown below. All the values shown below are values at a leaf temperature of 25°C. V cmax = 90.58 [μmol·m -2 ·s -1 J max = 154.99 [μmol·m -2 ·s -1 R d = 1.39 [μmol·m -2 ·s -1 K c = 404.9 [μmol·mol -1 K O = 278.4 [μmol·mol -1 Γ * = 42.75 [μmol·mol -1 θ = 0.7 φ = 0.36 g0 = 0.034 [mol·m -2 ·s -1 g1 = 4.43

[0147] The variable V cmax , J max , R d , K c , K O , Γ * are parameters with temperature dependence. The values of the variables V cmax , R d , K c , K O , Γ * are obtained by the Arrhenius equation based on the values at a leaf temperature of 25°C. The variable J max ​​​​​​​It is obtained by the modified Arrhenius equation based on the value when the leaf temperature is 25°C. Since the Arrhenius equation is well-known, detailed description in this specification is omitted. [Third Embodiment]

[0148] The configurations of the learning data creation device, machine learning device, and estimation device according to the third embodiment described below are the same as those of the learning data creation device, machine learning device, and estimation device according to the first embodiment unless otherwise specified, and thus duplicate explanations are omitted.

[0149] FIG. 13 is a schematic diagram for explaining an artificial neural network used in an estimation algorithm according to the third embodiment of the present invention.

[0150] In the third embodiment, a transpiration demand model is used as the plant physiological ecology model, and as environmental data, photosynthetic photon flux density (PPFD), atmospheric temperature T a , atmospheric humidity VPD a , and the leaf surface boundary layer conductance g aH for heat transport are used. The environmental data does not include the atmospheric CO2 concentration C a . However, the atmospheric CO2 concentration C a is measured and recorded together with the environmental data. The transpiration demand model used in the third embodiment is a model that does not use the atmospheric CO2 concentration C l in the calculation of the individual leaf transpiration rate Tr a . That is, in the third embodiment, although the atmospheric CO2 concentration C a is measured and recorded together with the environmental data, the atmospheric CO2 concentration C l is not used in the calculation of the individual leaf transpiration rate Tr a . The third embodiment is different from the first embodiment in that the atmospheric CO2 concentration C l is not used in the calculation of the individual leaf transpiration rate Tr a , and the atmospheric CO2 concentration C a is directly set in the input layer 51 of the estimation algorithm 5.

[0151] At the time of creating learning data, the individual leaf transpiration rate calculation unit 11 calculates the individual leaf transpiration rate Tr based on the environmental data 41 in the field 90 and the transpiration demand model. l The learning data creation unit 13 associates the transpiration rate per plant 43 of the plant 91 measured in the field 90 with the individual leaf transpiration rate Tr l , an index related to the leaves, and the CO2 concentration C a in the field 90 to create learning data 4. The learning data 4 is composed of the transpiration rate per plant 43 of the plant 91, the individual leaf transpiration rate Tr l , an index related to the leaves, and the CO2 concentration C a in the field 90, each of which is associated with each other.

[0152] At the time of machine learning, the learning unit 21 sets the individual leaf transpiration rate Tr l , an index related to the leaves, and the CO2 concentration C a in the field 90 in the input layer 51, sets the transpiration rate per plant 43 in the output layer 53, and learns the estimation algorithm 5.

[0153] At the time of estimation, the transpiration rate estimation unit 35 inputs the individual leaf transpiration rate Tr l calculated by the individual leaf transpiration rate calculation unit 33, the index related to the leaves calculated by the index calculation unit 34, and the CO2 concentration in the field 80 acquired using the sensing device 6 into the input layer 51 of the learned estimation algorithm 5, whereby an estimated value of the transpiration rate per plant of the plant 81 is output from the output layer 53.

[0154] According to the third embodiment of the present invention, the estimation algorithm 5 is learned by directly setting the CO2 concentration C l of the atmosphere in the input layer 51 of the estimation algorithm 5 without using the CO2 concentration C a of the atmosphere for calculating the individual leaf transpiration rate Tr, and the transpiration rate per plant of the plant 81 is calculated using the learned estimation algorithm 5. a [Other forms]

[0155] As described above, the present invention has been described with reference to specific embodiments, but the present invention is not limited to the above-described embodiments.

[0156] In each of the above-described embodiments, the learning of the estimation algorithm 5 is performed using the measurement data 43 of the transpiration rate per stock, and the transpiration rate per stock is estimated according to the learned estimation algorithm 5. However, the transpiration rate used for learning and the transpiration rate to be estimated are not limited to the transpiration rate per stock. The learning of the estimation algorithm 5 can also be performed using the measurement data of the transpiration rate of the community, and the transpiration rate of the community can be estimated according to this learned estimation algorithm 5. When the community is composed of a plurality of stocks, the transpiration rate of the community is the sum of the transpiration rates per stock for each of the plurality of stocks. Therefore, if the number of stocks constituting the community can be grasped, the transpiration rate per stock can be calculated from the transpiration rate of the community, and conversely, the transpiration rate of the community can be calculated from the transpiration rate per stock.

[0157] In each of the above-described embodiments, an artificial neural network is used for the configuration of the estimation algorithm 5, but the estimation algorithm 5 is not limited to an artificial neural network. As long as the estimation algorithm can be learned using learning data, various machine learning algorithms can be used for the estimation algorithm.

[0158] In each of the above-described embodiments, the index calculation unit 12 calculates both the leaf area index and the light reception efficiency index of the leaf as indices related to the leaf. However, the index to be calculated as an index related to the leaf used for creating the learning data 4 can be at least either the leaf area index or the light reception efficiency index of the leaf.

[0159] In each of the above-described embodiments, the learning data creation device 1 and the machine learning device 2 are configured as separate devices, but the learning data creation device 1 and the machine learning device 2 can be integrated and configured as one device.

[0160] In each of the above-described embodiments, the learning data creation device 1, the machine learning device 2, and the estimation device 3 are communicably connected to each other via the network 10. However, these learning data creation device 1, machine learning device 2, and estimation device 3 can be connected so as to be able to exchange data via a recording medium such as a DVD-ROM or a memory card, for example.

[0161] In each of the above-described embodiments, the administrator of the farmland 90 uses the learning data creation device 1 and the machine learning device 2. However, the person handling the learning data creation device 1 and the machine learning device 2 is not limited to the administrator of the farmland 90. For example, the administrator of the farmland 90 acquires the data 41, 42, 43 in the farmland 90, and a person proficient in machine learning technology, such as a data scientist, can create the learning data and perform machine learning of the estimation algorithm.

[0162] In each of the above-described embodiments, the learning data creation device 1 is realized as an integrated device. However, the learning data creation device 1 does not have to be an integrated device, and a processor, a main storage device, an auxiliary storage device 19, etc. may be arranged separately, and these may be communicably connected to each other via a network. The same applies to the machine learning device 2 and the estimation device 3. Also, with respect to the sensing device 6, the imaging device 7, the display device 8, and the input device 9 connected to the estimation device 3, these do not have to be arranged in one place, and each may be arranged separately and communicably connected to each other via a network.

[0163] In each of the above-described embodiments, each functional block of the learning data creation device 1 is executed by a single processor. However, these functional blocks do not have to be executed by a single processor, and may be executed distributively by a plurality of processors. The same applies to the machine learning device 2 and the estimation device 3.

[0164] Each functional block of the learning data creation device 1, the machine learning device 2, and the estimation device 3 may be partially or entirely cloudified in a server device (not shown) connected via the network 10. [Embodiment]

[0165] Examples of the present invention are shown below to clarify the features of the present invention.

Example

[0166] In Example 1, the validity of the transpiration rate per stock estimated by the estimation device was verified. All the data used for the verification was obtained in a field for test research. The plants cultivated in the field for test research were eggplants in this example.

[0167] The verification was performed in the following procedure. First, using a set of transparent open-type chambers installed in the field for test research, various data used for validity verification was obtained. The obtained data was the environmental data of the field, the directly downward view image data of the plants (eggplants) cultivated in the field, and the actual measured values of the transpiration rate per stock of the plants. These various data were obtained over a total of 13 days in the field for test research in accordance with the modes shown in Steps S1 to S3 of the first embodiment described above. Thereafter, the combination of the obtained environmental data, the directly downward view image data, and the actual measured values of the transpiration rate per stock was managed as a data set.

[0168] Next, the obtained multiple data sets for 13 days were randomly divided at a predetermined ratio. In this example, the obtained multiple data sets were randomly divided at a ratio of training:validation:test = 0.4:0.1:0.5.

[0169] Next, using the environmental data and the directly-viewed plant image data included in the training dataset and the validation dataset, learning data was created by the learning data creation device. This learning data was created in accordance with the mode shown in steps S3 to S7 of the first embodiment described above. Next, using the learning data created in this way, the estimation algorithm was learned by the machine learning device in accordance with the mode shown in step S8 of the first embodiment described above. The training dataset and the validation dataset used for creating the learning data and learning the estimation algorithm corresponded to approximately 50% of the acquired dataset.

[0170] Next, in order to verify the validity of the estimated value obtained by the estimation device, it was verified whether the estimation device could reproduce the test dataset using this learned estimation algorithm and the estimation device. The test dataset corresponded to the remaining approximately 50% of the acquired dataset.

[0171] First, using the environmental data and the directly-viewed plant image data included in the test dataset, the transpiration rate per stock was estimated by the estimation device. This estimation was performed in accordance with the mode shown in steps S13 to S15 of the first embodiment described above. Next, the reproducibility of the estimated value by the estimation device was confirmed by comparing the estimated value of the transpiration rate per stock obtained by the estimation device with the actual measured value of the transpiration rate per stock included in the test dataset used for the estimation of the estimated value.

[0172] FIG. 14 is a graph for verifying the validity of the transpiration rate per stock estimated by the estimation device in Example 1. (A) is a graph showing the variation of the estimated transpiration rate per stock together with the measured values for each weather. (B) is a graph showing the correlation between the estimated value and the measured value of the transpiration rate per stock.

[0173] As shown in (A), the estimated value and the measured value generally agreed. Also, regarding the correlation between the estimated value and the measured value, as shown in (B), the coefficient of determination R 2 is R2 It was 0.95, and the Root Mean Square Error (RMSE) was RMSE = 0.06.

[0174] As a result, it was confirmed that the estimated value by the estimation device faithfully reproduced the influence of the weather and was a highly valid value.

Example

[0175] In Example 2, the validity of the Leaf Area Index (LAI) introduced as an index related to leaves in the present invention was verified.

[0176] The verification was performed according to the following procedure. In the field, pruning or leaf removal operations are performed when harvesting plants. Each time such pruning or leaf removal operations were performed in the field, the Leaf Area Index (LAI) of the plant stocks was calculated. At that time, the number of flowers and fruits of the plants for the stocks for which the Leaf Area Index (LAI) was calculated was recorded. By dividing the recorded number of flowers and fruits by the area of the ground occupied by the plant stocks in a directly downward view, the number of flowers and fruits per unit area was calculated. The plants cultivated in the field were eggplants in this example.

[0177] FIG. 15 is a graph showing side by side on the same time axis the seasonal variations in the Leaf Area Index and the seasonal variations in the number of flowers and fruits in Example 2.

[0178] As shown in FIG. 15, the variation in the Leaf Area Index shown in the upper graph generally reproduced the variation in the number of flowers and fruits shown in the lower graph. That is, the Leaf Area Index was able to generally reproduce the change in leaf area due to pruning or leaf removal operations accompanying harvesting. As a result, it was confirmed that in the present invention, the Leaf Area Index introduced into the input layer of the estimation algorithm as an index related to leaves is input data with high explanatory power for the output layer.

Example

[0179] In Example 3, the performance of the estimation algorithm for estimating the transpiration rate was verified. The estimation algorithm was configured using an artificial neural network, and four types shown in FIGS. 16(A) to 16(D) were constructed according to the types of data set in the input layer of the artificial neural network. The verification was performed by comparing the estimated value by the estimation device with the actual measured value for each model shown in FIGS. 16(A) to 16(D) to confirm the reproducibility of the estimated value by the estimation device, and by comparing the reproducibility of the estimated values among the models.

[0180] FIG. 16 is a schematic diagram for explaining the artificial neural network used for verifying the performance of the estimation algorithm in Example 3.

[0181] (A) is a comparative example, in which environmental data is directly set in the input layer of the artificial neural network without calculating the transpiration rate of individual leaves from the environmental data and setting it in the input layer of the artificial neural network. The comparative example of (A) is called a simple model.

[0182] (B) is an example of the present invention and corresponds to the first embodiment. In the example of (B), using the evapotranspiration demand (ED) model as a plant physiological ecology model, the transpiration rate of individual leaves is calculated from environmental data, and the calculated transpiration rate of individual leaves is set in the input layer of the artificial neural network. The environmental data does not include the atmospheric CO2 concentration C a . The example of (B) is called the ED model.

[0183] (C) is an example of the present invention and corresponds to the second embodiment. In the example of (C), using the FvCB model as a plant physiological ecology model, the transpiration rate of individual leaves is calculated from environmental data, and the calculated transpiration rate of individual leaves is set in the input layer of the artificial neural network. The environmental data includes the atmospheric CO2 concentration C a . The example of (C) is called the FvCB model.

[0184] (D) is an example of the present invention and corresponds to the third embodiment. In the example of (D), the evapotranspiration demand (ED) model is used as the plant physiological and ecological model to calculate the transpiration rate of each leaf from environmental data, and the calculated transpiration rate of each leaf is set in the input layer of the artificial neural network. The CO2 concentration C a of the atmosphere is not included in the environmental data and is directly set in the input layer of the artificial neural network. The example of (D) is called the ED + CO2 model.

[0185] In each of the models shown in FIGS. 16(A) to (D), learning data was created and machine learning was performed using the measured data of the transpiration rate of the community instead of the measured data of the transpiration rate per plant. Similarly, at the time of estimation, an estimated value of the transpiration rate of the community was obtained using the learned artificial neural network.

[0186] In Example 3, the reproducibility of the estimated value was confirmed in the same manner as in Example 1. That is, a plurality of data sets obtained over a plurality of days were randomly divided at a ratio of training: validation: test = 0.4: 0.1: 0.5, and learning data was created, the estimation algorithm was learned, and estimation was performed by the estimation device. A list of the data sets used for constructing the artificial neural network in Example 3 is shown in FIG. 17.

[0187] FIGS. 18 to 20 are diagrams for comparing the reproducibility of the estimated values among the models in Example 3. In FIGS. 18 to 20, graphs showing the correlation between the estimated value and the measured value of the transpiration rate of the community are arranged in a grid of 6 vertical and 6 horizontal according to the period during which the data set was obtained. Among the plurality of different models represented by the points of different plots in the graphs showing the correlation between the estimated value and the measured value shown in these figures, the model whose plot points are along the 1: 1 line indicated by the solid line has a high reproducibility of the estimated value.

[0188] Figure 18 is a diagram for comparing the reproducibility of estimated values between the FvCB model and the simple model in Example 3. As shown in Figure 18, in the comparison between the FvCB model and the simple model, it was confirmed that the FvCB model clearly has higher reproducibility of estimated values than the simple model.

[0189] Figure 19 is a diagram for comparing the reproducibility of estimated values between the ED model and the simple model in Example 3. As shown in Figure 19, in the comparison between the ED model and the simple model, it was confirmed that the ED model has higher reproducibility of estimated values than the simple model. Also, by comparing Figure 18 and Figure 19, when comparing the FvCB model and the ED model with the simple model as a reference, it was confirmed that there is no inferiority in the reproducibility of estimated values between these models, and the reproducibility of estimated values is sufficient even for the ED model.

[0190] Figure 20 is a diagram for comparing the reproducibility of estimated values between the ED+CO2 model and the simple model in Example 3. As shown in Figure 20, in the comparison between the ED+CO2 model and the simple model, it was confirmed that the ED+CO2 model has higher reproducibility of estimated values than the simple model. Also, by comparing Figure 19 and Figure 20, when comparing the ED model and the ED+CO2 model with the simple model as a reference, it was confirmed that the ED+CO2 model has a lower reproducibility of estimated values than the ED model.

[0191] Among the models shown in Figures 16(A) to (D), for the ED model shown in (B), the reproducibility of estimated values was further confirmed for the case where environmental factors differ significantly between the learning of the estimation algorithm and the estimation by the estimation device. As a comparative example, for the simple model shown in (A), the reproducibility of estimated values was also confirmed in the same manner as the ED model shown in (B).

[0192] Figure 21 is a graph showing the results of verifying the performance of the estimation algorithm in Example 3 for the case where environmental factors differ significantly between the learning of the estimation algorithm and the estimation by the estimation device.

[0193] (A) shows the result of training the estimation algorithm using the dataset "Oct-3" shown in FIG. 17 as the training dataset and performing estimation by the estimation device using the dataset "May-1" shown in FIG. 17 as the test dataset.

[0194] (B) shows the result of training the estimation algorithm using the dataset "May-3" shown in FIG. 17 as the training dataset and performing estimation by the estimation device using the dataset "Oct-1" shown in FIG. 17 as the test dataset.

[0195] As shown in (A) and (B) respectively, for the ED model, even when the environmental factors were significantly different between the learning time and the estimation time, the estimated values and the measured values generally agreed. In contrast, for the simple model, when the environmental factors were significantly different between the learning time and the estimation time, the estimated values and the measured values were greatly deviated.

[0196] Thus, for the ED model shown in FIG. 16(B), even when the environmental factors were significantly different between the learning of the estimation algorithm and the estimation by the estimation device, the estimated value by the estimation device faithfully reproduced the measured value, and it was confirmed that the value was highly valid.

Explanation of Signs

[0197] 1 Learning data creation device 2 Machine learning device 3 Estimation device 4 Learning data 5 Estimation algorithm 6 Sensing device (sensor) 7 Imaging device (digital camera) 8 Display device 9 Input device 10 Network 11 Individual leaf transpiration rate calculation unit 12 Index calculation unit 121 Leaf area extraction unit 122 Leaf area index calculation unit 123 Light reception efficiency index calculation unit 13 Learning data creation unit 19 Auxiliary storage device 21 Learning unit 29 Auxiliary storage device 31 Environmental data acquisition unit 32 Plant image acquisition unit 33 Transpiration rate per individual leaf calculation unit 34 Index calculation unit 341 Leaf area extraction unit 342 Leaf area index calculation unit 343 Light reception efficiency index calculation unit 35 Transpiration rate estimation unit 39 Auxiliary storage device 41 Environmental data 42 Image data 43 Measurement data of transpiration rate per plant 51 Input layer 52 Intermediate layer 53 Output layer 80 Producer's field 81 Plant 90 Experimental research field 91 Plant 92 Chamber 93 Intake port 94 Exhaust port 95 Gas analyzer 96 Flow meter 97 PPFD sensor 98a, 98b Thermocouple 98c Anemometer 99 Imaging device (digital camera) 100 Transpiration rate estimation support system

Claims

1. A learning data creation device for creating learning data used for learning an estimation algorithm for estimating the transpiration rate of a plant, comprising: An individual leaf transpiration rate calculation unit that calculates the individual leaf transpiration rate based on environmental data in a field; An index calculation unit that calculates an index related to the leaf based on an image including the leaf of the plant cultivated in the field; A learning data creation unit that creates learning data by associating the transpiration rate of the plant measured in the field with the individual leaf transpiration rate and the index related to the leaf; Comprising; The environmental data includes photosynthetic photon flux density, ambient air temperature, ambient humidity, and leaf surface boundary layer conductance for heat transport; The individual leaf transpiration rate calculation unit calculates the individual leaf transpiration rate based on the environmental data and a plant physiological and ecological model; The index related to the leaf includes a light reception efficiency index of the leaf; The index calculation unit includes a light reception efficiency index calculation unit that calculates the light reception efficiency index by integrating the relative luminance of the leaf region in the image. A learning data creation device.

2. The learning data creation device according to claim 1, wherein the index calculation unit includes a leaf region extraction unit that extracts the leaf region in the image by image analysis using artificial intelligence.

3. The learning data creation device according to any one of claims 1 or 2, wherein the index related to the leaf further includes a leaf area index of the leaf.

4. The learning data creation device according to claim 3, wherein the index calculation unit includes a leaf area index calculation unit that calculates the leaf area index using a ratio occupied by a region other than the leaf in the image obtained by binarizing the image.

5. The plant physiological and ecological model includes a transpiration demand model; The learning data creation device according to claim 1, wherein the individual leaf transpiration rate calculation unit calculates the individual leaf transpiration rate based on the environmental data and the transpiration demand model.

6. The plant physiological and ecological model includes an FvCB model; The environmental data further includes an ambient carbon dioxide concentration; The learning data creation device according to claim 1, wherein the individual leaf transpiration rate calculation unit calculates the individual leaf transpiration rate based on the environmental data and the FvCB model.

7. A machine learning unit that performs supervised machine learning of an estimation algorithm for estimating the transpiration rate of the plant, using the learning data created by the learning data creation device according to any one of claims 1 to 6 as teacher data. A machine learning device comprising the same.

8. The machine learning device according to claim 7, wherein the estimation algorithm is configured using an artificial neural network.

9. An environmental data acquisition unit that acquires environmental data in the field, A plant image acquisition unit that acquires an image including leaves of a plant cultivated in the field, An individual leaf transpiration rate calculation unit that calculates an individual leaf transpiration rate based on the environmental data, An index calculation unit that calculates an index related to the leaf based on the image, A transpiration rate estimation unit that estimates the transpiration rate of the plant based on the individual leaf transpiration rate and the index related to the leaf according to the estimation algorithm learned by the machine learning device according to claim 7 or 8. An estimation device comprising the same.

10. A learning data creation method for creating learning data used for learning an estimation algorithm for estimating the transpiration rate of a plant, An individual leaf transpiration rate calculation step of calculating an individual leaf transpiration rate based on environmental data in the field, An index calculation step of calculating an index related to the leaf based on an image including leaves of a plant cultivated in the field, A learning data creation step of creating learning data by associating the transpiration rate of the plant measured in the field with the individual leaf transpiration rate and the index related to the leaf, Including, The environmental data includes photosynthetic photon flux density, ambient air temperature, ambient humidity, and leaf surface boundary layer conductance for heat transport. The individual leaf transpiration rate calculation step calculates the individual leaf transpiration rate based on the environmental data and a plant physiological ecology model. The index related to the leaf includes a light reception efficiency index of the leaf. The index calculation step includes a light reception efficiency index calculation step of calculating the light reception efficiency index by integrating the relative luminance of the leaf region in the image. A learning data creation method.

11. A machine learning step of performing supervised machine learning of an estimation algorithm for estimating the transpiration rate of the plant, using the learning data created by the learning data creation method according to claim 10 as teacher data. Including, a machine learning method.

12. An environmental data acquisition step of acquiring environmental data in a field; A plant image acquisition step of acquiring an image including leaves of a plant cultivated in the field; An individual leaf transpiration rate calculation step of calculating an individual leaf transpiration rate based on the environmental data; An index calculation step of calculating an index related to the leaf based on the image; A transpiration rate estimation step of estimating the transpiration rate of the plant based on the individual leaf transpiration rate and the index related to the leaf according to an estimation algorithm learned by the machine learning method according to claim 11; An estimation method comprising:

13. A program for causing a computer to execute each step of the learning data creation method according to claim 10.

14. A program for causing a computer to execute each step of the machine learning method according to claim 11.

15. A program for causing a computer to execute each step of the estimation method according to claim 12. ​ ​

Images