Integrated water stress value estimation method, integrated water stress value estimation device, program, training method for machine model, and trained machine model

Abstract

To extract features of fruit using a machine model and then estimate an integrated water stress value from temporal change of features.SOLUTION: An integrated water stress value estimation method (S1) includes: an acquisition step (S11) of imaging one fruit that a tree bears at a first time point and a second time point to acquire two images differing in the time points of imaging; an input step (S12) of inputting the two acquired images to a machine model which is so trained as to output an estimated integrated water stress value when the two images obtained by imaging the one fruit respectively at the first time point and the second time point are input; and an output step (S13) of outputting the estimated integrated water stress value that the machine model outputs.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an integrated water stress value estimation method, an integrated water stress value estimation device, a program, a machine model learning method, and a learned machine model.

Background Art

[0002] A technique for estimating the integrated water stress value of a fruit tree from the temporal change in the size of a fruit is known as a conventional technique.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the conventional technology as described above, depending on weather conditions and the like, there may be no correlation between the degree of the temporal change in the size of the fruit and the estimated integrated water stress value. In such a case, there is a problem that the accuracy of the integrated water stress value estimated from the temporal change in the size of the fruit decreases.

[0005] One aspect of the present invention aims to realize a technique for extracting fruit characteristics using a machine model and estimating an integrated water stress value from the temporal change in the characteristics.

Means for Solving the Problems

[0006] In order to solve the above problems, an integrated water stress value estimation method according to one aspect of the present invention includes: for one fruit in a state of being set on a tree, imaging is performed at a first time point and a second time point, and two images with different imaging time points are acquired; an acquisition step; and the two acquired images are input into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, for one fruit are input, an input step; and an output step of outputting the estimated integrated water stress value output by the machine model.

[0007] In order to solve the above problems, an integrated water stress value estimation device according to one aspect of the present invention includes: an acquisition unit that performs imaging on one fruit in a state of being set on a tree at a first time point and a second time point, and acquires two images with different imaging time points; an input unit that inputs the two acquired images into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, for one fruit are input; and an output unit that outputs the estimated integrated water stress value output by the machine model.

[0008] In order to solve the above problems, a program according to one aspect of the present invention causes at least one processor to execute: for one fruit in a state of being set on a tree, an acquisition step of performing imaging at a first time point and a second time point, and acquiring two images with different imaging time points; an input step of inputting the two acquired images into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, for one fruit are input; and an output step of outputting the estimated integrated water stress value output by the machine model.

[0009] In order to solve the above problems, a method for learning a machine model according to an aspect of the present invention includes, for one fruit in a state of being fruited on a tree, imaging at a first time point and a second time point, two images with different imaging time points, and an estimated integrated water stress value in a period between the first time point and the second time point estimated from a state of the tree on which the fruit is fruited, and acquiring a plurality of sets of the set data as teacher data; and a learning step of learning to output an estimated integrated water stress value when the two images, i.e., the image taken at the first time point and the image taken at the second time point, are input for one fruit using the teacher data.

[0010] In order to solve the above problems, a learned machine model according to an aspect of the present invention takes two images with different imaging time points, which are obtained by imaging one fruit in a state of being fruited on a tree at a first time point and a second time point, as inputs, and outputs an estimated integrated water stress value.

Advantages of the Invention

[0011] According to an aspect of the present invention, it is possible to extract features of a fruit using a machine model and estimate an integrated water stress value from a change over time of the features.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Mode for Carrying Out the Invention

[0013] 〔Embodiment 1〕 Hereinafter, an embodiment of the present invention will be described in detail.

[0014] (Outline of the integrated water stress value estimation device 1) In fruits where there is a positive correlation between the integrated water stress in summer and the sugar content increase, it is known that by applying drought stress (water stress) in summer, fruits with high sugar content required by the market can be produced. Specific examples of such fruits include citrus fruits such as unshu mikan, or pears. On the other hand, excessive drought stress may cause high oxidation of fruits and a decline in the tree vigor of fruit trees, which may inhibit the stable production of fruits. Therefore, it is necessary to maintain an appropriate level of water stress. Although the pressure chamber method can be mentioned as a method for investigating the water stress value, it is necessary to conduct the investigation before dawn, and the equipment used is also expensive, so it has not been popularized among producers. Therefore, the development of a simple water stress value estimation method in the field is required.

[0015] The integrated water stress value estimation device 1 according to the present embodiment is a device for extracting the characteristics of fruits using a mechanical model and estimating the integrated water stress value from the temporal change of the characteristics.

[0016] (Configuration of the integrated water stress value estimation device 1) The configuration of the integrated water stress value estimation device 1 according to the present embodiment will be described. FIG. 1 is a block diagram showing the configuration of the integrated water stress value estimation device 1 according to the present embodiment. The integrated water stress value estimation device 1 includes an acquisition unit 11, an input unit 12, an output unit 13, and a mechanical model 14.

[0017] (Acquisition unit 11) The acquisition unit 11 captures images of one fruit in a state of being set on a tree at a first time point and a second time point, and acquires two images with different imaging time points. The period of the imaging interval may be, for example, about several days to several weeks, or may be from 7 days to 10 days. Each image may include an identification image for identifying the fruit. However, when inputting the images into the machine model 14, the identification image for identifying the fruit may be erased.

[0018] (Input unit 12) The input unit 12 inputs the two acquired images into a machine model 14 that has been trained to output an estimated integrated water stress value when two images, an image captured at the first time point and an image captured at the second time point, are input for one fruit. The description of the machine model 14 will be described later.

[0019] (Output unit 13) The output unit 13 outputs the estimated integrated water stress value output by the machine model 14. The output destination may be, for example, various output devices such as a display device (display, etc.), a printing device (printer, etc.), a speaker, various storage devices, or other systems such as an irrigation system.

[0020] (Machine model 14) The machine model 14 is a machine model that has been trained to output an estimated integrated water stress value when two images, an image captured at the first time point and an image captured at the second time point, are input for one fruit.

[0021] The mechanical model 14 may be, for example, a mechanical model using a neural network. Specific examples of the neural network include a convolutional neural network (CNN), a recurrent neural network (RNN), and a fully connected neural network (FCNN).

[0022] The integrated water stress value estimation device 1 extracts a feature amount including the features of the fruit in the image using the mechanical model 14. The components of the feature amount may be, for example, the size of the fruit, the skin color, the texture of the fruit, and the like. By using the feature amount including such a plurality of components, the mechanical model 14 can estimate the integrated water stress value with higher accuracy.

[0023] Further, for example, the estimated integrated water stress value output by the mechanical model 14 may be the estimated integrated water stress value of the tree on which the fruit included in the two input images has set.

[0024] (Effect of the integrated water stress value estimation device 1) As described above, in the integrated water stress value estimation device 1, for one fruit in the state of being set on a tree, imaging is performed at a first time point and a second time point, and an acquisition unit that acquires two images with different imaging time points, and the two acquired images are input into a mechanical model that is trained to output an estimated integrated water stress value when two images, i.e., the image taken at the first time point and the image taken at the second time point, for one fruit are input. The integrated water stress value estimation device 1 includes an input unit and an output unit that outputs the estimated integrated water stress value output by the mechanical model.

[0025] According to the integrated water stress value estimation device 1 configured as described above, the features of the fruit can be extracted using the mechanical model, and the integrated water stress value can be estimated from the temporal change of the features.

[0026] (Flow of the integrated water stress value estimation method S1) The flow of the integrated water stress value estimation method S1 executed by the integrated water stress value estimation device 1 will be described. FIG. 2 is a flowchart showing an example of the flow of the integrated water stress value estimation method S1 according to the present embodiment. As shown in FIG. 2, the integrated water stress value estimation method S1 includes steps S11 to S13. Note that FIG. 2 also describes a step S14 of inputting the output estimated integrated water stress value into the irrigation system as an example of a method of using the estimated integrated water stress value. The irrigation system is a system that automatically irrigates fruit trees based on the value of the estimated integrated water stress value. Thus, the estimated integrated water stress value obtained by the integrated water stress value estimation method S1 can be used to determine whether to irrigate the tree on which the fruit is set.

[0027] (Step S11) In step S11, the acquisition unit 11 captures images of one fruit in the state of being set on the tree at a first time point and a second time point, and acquires two images with different imaging time points.

[0028] (Step S12) In step S12, the input unit 12 inputs the two acquired images into the machine model 14 that has been learned to output the estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, of one fruit are input. The machine model 14 may be, for example, a machine model using a neural network.

[0029] (Step S13) In step S13, the output unit 13 outputs the estimated integrated water stress value output by the machine model 14. For example, the estimated integrated water stress value output by the machine model 14 may be the estimated integrated water stress value of the tree on which the fruit included in the two input images is set.

[0030] (Step S14) In step S14, for example, the integrated water stress value estimation device 1 may input the output estimated integrated water stress value into the irrigation system.

[0031] (Effect of the integrated water stress value estimation method S1) As described above, in the integrated water stress value estimation method S1, for one fruit in the fruiting state on the tree, imaging is performed at a first time point and a second time point, and an acquisition step of acquiring two images with different imaging time points, and the two acquired images are input into a machine model learned to output an estimated integrated water stress value when two images, i.e., an image captured at the first time point and an image captured at the second time point, for one fruit are input. The input step and the output step of outputting the estimated integrated water stress value output by the machine model are adopted.

[0032] According to the integrated water stress value estimation method S1 configured as described above, the characteristics of the fruit can be extracted using the machine model, and the integrated water stress value can be estimated from the temporal change of the characteristics.

[0033] (Configuration of the integrated water stress value estimation device 1A and the machine model 2) The configuration of the integrated water stress value estimation device 1A according to the present embodiment will be described. FIG. 3 is a block diagram showing a configuration example of the integrated water stress value estimation device 1A according to the present embodiment. In the example of FIG. 3, the integrated water stress value estimation device 1A includes an acquisition unit 11, an input unit 12, an output unit 13, and a communication unit 15. The main difference between the integrated water stress value estimation device 1 and the integrated water stress value estimation device 1A is that the integrated water stress value estimation device 1 has the machine model 14 inside, while the integrated water stress value estimation device 1A is connected to an external machine model 2 via the network 5. It should be noted that members having the same functions as the members described in the items of the integrated water stress value estimation device 1 are also denoted by the same reference numerals in the items of the integrated water stress value estimation device 1A.

[0034] (Database 100) For example, as shown in FIG. 3, the integrated water stress value estimation device 1A may be connected to an external database 100. In the example of FIG. 3, the database 100 stores the data of the images acquired by the acquisition unit 11. Also, in the example of FIG. 3, the acquisition unit 11 captures images of one fruit in a state of bearing fruit on a tree at a first time point and a second time point, and acquires, from the database 100 external to the integrated water stress value estimation device 1A, a first image and a second image, which are a pair of two images with different imaging time points.

[0035] (Pre-treatment of the fruit) For example, as shown in the first image and the second image of FIG. 3, a small piece of plastic or the like serving as a distance indicator may be attached to the apex of the fruit, and then the image of the fruit may be input into a machine model. Also, for example, identification information such as a QR code (registered trademark) may be attached to the small piece and used for identification between images. Specific examples of the identification content include the identification of the same fruit between different images. Note that in the examples of the first image and the second image of FIG. 3, the display positions of the QR codes (registered trademarks) that are not necessary for estimating the integrated water stress value are painted black.

[0036] (Output destination of the result data of the integrated water stress value estimation device 1A) Also, as shown in the example of FIG. 3, various result data output from the integrated water stress value estimation device 1A may be input into an external output device 3 and an irrigation system 4. Specific examples of the result data include the estimated integrated water stress value.

[0037] (Communication unit 15) The communication unit 15 communicates with, for example, a device external to the integrated water stress value estimation device 1A. For example, the communication unit 15 may communicate with the mechanical model 2 connected to the integrated water stress value estimation device 1A via the network 5. Further, for example, the input unit 12 may input the two images acquired by the acquisition unit 11 to the mechanical model 2 via the communication unit 15 and the network 5. Note that the specific configuration of the network 5 does not limit this embodiment, but as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0038] (Mechanical model 2) Similar to the mechanical model 14, the mechanical model 2 is a mechanical model that has been trained to output an estimated integrated water stress value when two images, an image captured at a first time point and an image captured at a second time point, are input for one fruit.

[0039] For example, the mechanical model 2 may include a first mechanical model 21 that takes each of the two images as an input and outputs a matrix for each of the two images, where the number of rows and columns of the two matrices are the same.

[0040] In FIG. 3, the first mechanical model 21 may take, for example, a first image that is an image captured at a first time point and a second image that is an image captured at a second time point as inputs, and output a matrix for each of the first image and the second image. Further, the matrix output by the first mechanical model 21 may indicate a feature amount including the features of the fruit in the image, as described in the items of the mechanical model 14. In the example of FIG. 3, the matrix for the first image is referred to as the first matrix, and the matrix for the second image is referred to as the second matrix. Note that the number of rows and columns are the same between the first matrix and the second matrix.

[0041] Further, for example, the machine model 2 may include a second machine model 22 that takes as input the difference between the first matrix and the second matrix, which are two output matrices, and outputs an estimated integrated water stress value.

[0042] For example, since the integrated water stress value is the integrated value of the water stress values during a predetermined period, when directly inputting the difference between the first matrix and the second matrix to the second machine model 22, it is necessary to always keep the same the order of the imaging times of the first image and the second image, and the order of taking the difference between the first matrix and the second matrix. However, for example, when inputting the absolute value of the difference between the first matrix and the second matrix to the second machine model 22, the order of the imaging times of the first image and the second image may be reversed.

[0043] The machine model 2 may be, for example, a machine model using a neural network, similar to the machine model 14. As a specific example in the components of the machine model 2, a CNN may be used for the first machine model 21 and an FCNN may be used for the second machine model 22, respectively.

[0044] Further, for example, the estimated integrated water stress value output by the machine model 2 may be the estimated integrated water stress value of the tree on which the fruits included in the two input images are set. That is, in the example of FIG. 3, the estimated integrated water stress value output by the machine model 2 may be the estimated integrated water stress value of the tree on which the fruits included in the input first image and second image are set.

[0045] Further, the machine model 2 may output the estimated integrated water stress value to the integrated water stress value estimation device 1A via, for example, the network 5 and the communication unit 15. Further, the output unit 13 may output the estimated integrated water stress value, for example, as the result data of the integrated water stress value estimation device 1A to the external output device 3 and the irrigation system 4.

[0046] (Output device 3) The output device 3 is configured to include at least one of output devices such as a display and a touch panel. Further, the output device 3 may output information related to the input estimated integrated water stress value via characters, voice, or images, for example. The information related to the estimated integrated water stress value output in this way may be used, for example, to determine the necessity of irrigation for fruit trees.

[0047] (Irrigation system 4) As described in the item of the integrated water stress value estimation method S1, the irrigation system 4 is, for example, a system that automatically irrigates fruit trees based on the value of the estimated integrated water stress value. For example, in this way, the estimated integrated water stress value may be used to determine whether to irrigate the tree on which the fruit is set.

[0048] (Effect of the integrated water stress value estimation device 1A and the mechanical model 2) According to the integrated water stress value estimation device 1A and the mechanical model 2 configured as described above, the same effects as those of the integrated water stress value estimation device 1 are achieved. Further, according to the integrated water stress value estimation device 1A and the mechanical model 2 configured as described above, by connecting the integrated water stress value estimation device 1A and the mechanical model 2 via the network 5, the mechanical model 2 among the components can be centrally managed by the server, and the processing can be distributed to improve the processing efficiency of the entire system.

[0049] (Configuration of the integrated water stress value estimation device 1A and the mechanical model 2A) FIG. 4 is a block diagram showing a configuration example of the integrated water stress value estimation device 1A according to the present embodiment. The configuration example in FIG. 4 is obtained by replacing the mechanical model 2 in the configuration example in FIG. 3 with the mechanical model 2A, and the other configurations are the same as those in FIG. 3.

[0050] (Mechanical model 2A) The mechanical model 2A is a mechanical model that is learned to output the estimated integrated water stress value when two images, an image captured at a first time point and an image captured at a second time point, are input for one fruit, similar to the mechanical model 2.

[0051] For example, the machine model 2A may include a first machine model that takes each of two images as input and outputs a matrix related to each of the two images, where the number of rows and columns of the two matrices are the same. Also, for example, the first machine model may include a first A machine model 21A that takes one of the two images, i.e., the image captured at the first time point and the image captured at the second time point, as input and outputs a matrix with a predetermined number of rows and columns, and a first B machine model 21B that takes the other of the two images, i.e., the image captured at the first time point and the image captured at the second time point, as input and outputs a matrix with a predetermined number of rows and columns.

[0052] In FIG. 4, the first A machine model 21A may, for example, take the first image as input and output a first matrix that is a matrix with a predetermined number of rows and columns. Also, in FIG. 4, the first B machine model 21B may, for example, take the second image as input and output a second matrix that is a matrix with a predetermined number of rows and columns. It is assumed that the number of rows and columns are the same between the first matrix and the second matrix.

[0053] Also, for example, the machine model 2A may include a second machine model 22 that takes the difference between the output first matrix and the second matrix as input and outputs an estimated integrated water stress value.

[0054] That is, the difference between the machine model 2 and the machine model 2A is that in the machine model 2, two images are input into one machine model, and two matrices related to each of the two input images are output from the one machine model, while in the machine model 2A, two images are input into separate machine models, and matrices related to each image are output separately from each machine model. Note that other configurations of the machine model 2A are the same as those of the machine model 2.

[0055] The mechanical model 2A may be, for example, a mechanical model using a neural network, similar to the mechanical model 14 and the mechanical model 2. As a specific example of the components of the mechanical model 2A, a CNN may be used for the first mechanical model 21A and the first B mechanical model 21B, and an FCNN may be used for the second mechanical model 22, respectively.

[0056] (Effects of the integrated water stress value estimation device 1A and the mechanical model 2A) According to the integrated water stress value estimation device 1A and the mechanical model 2A configured as described above, the same effects as those of the integrated water stress value estimation device 1A and the mechanical model 2 are achieved. Further, according to the integrated water stress value estimation device 1A and the mechanical model 2A configured as described above, by processing two images with separate mechanical models, the processing can be dispersed to further improve the processing efficiency of the entire system.

[0057] (Flow of the learning method S2 of the mechanical model) The flow of the learning method S2 of the mechanical model, which is executed when the mechanical models 14, 2, and 2A according to the present embodiment are learned, will be described. FIG. 5 is a flowchart showing an example of the flow of the learning method S2 of the mechanical model according to the present embodiment. In the description of the learning method S2 of the mechanical model, the mechanical models 14, 2, and 2A are collectively referred to as the "mechanical model".

[0058] (Step S21) In step S21, the mechanical model captures an image of one fruit in a state of being fruited on a tree at a first time point and a second time point, and acquires a plurality of sets of teacher data as a set of two images with different imaging time points and an estimated integrated water stress value in the period between the first time point and the second time point estimated from the state of the tree on which the fruit is fruited.

[0059] Generally, the degree of water stress of a tree body is represented by the water potential of leaves, and the estimated integrated water stress value may be evaluated by, for example, the following formula.

Equation

[0060] Regarding c in the above formula, the predetermined value for July - September may be, for example, -0.4 MPa.

[0061] (Step S22) In step S22, the machine model learns to output the estimated integrated water stress value when two images, an image captured at the first time point and an image captured at the second time point, are input for one fruit using the teacher data. When the learning image includes a QR code (registered trademark), it is preferable to use an image with the display area of the QR code (registered trademark) painted black for learning to prevent overfitting.

[0062] (Correlation between measured value and predicted value by machine model) After distributing the fruit images to the training data, validation data, and test data respectively, the results of analyzing the machine model will be described using the examples in FIGS. 6 and 7. In the examples of FIGS. 6 and 7, the machine model learned using the training data is analyzed using the validation data and test data. Note that in the examples of FIGS. 6 and 7, resnet50, a type of CNN, is used as the machine model.

[0063] Figure 6 shows an example of the analysis result of the validation dataset for the mechanical model. In the example of Figure 6, for the integrated water stress value SΨ, the predicted value output by the mechanical model when the validation data is input is compared with the measured value. In the example of Figure 6, when the correlation coefficient is R, the coefficient of determination is R 2 = 0.48, indicating a correlation between the predicted value and the measured value.

[0064] Figure 7 shows an example of the analysis result of the test dataset for the mechanical model. In the example of Figure 7, for the integrated water stress value SΨ, the predicted value output by the mechanical model when the test data is input is compared with the measured value. In the example of Figure 7, when the correlation coefficient is R, the coefficient of determination is R 2 = 0.44, indicating a correlation between the predicted value and the measured value.

[0065] (Trained mechanical model) The trained mechanical model according to this embodiment performs imaging on one fruit in a fruit-bearing state on a tree at a first time point and a second time point, takes two images with different imaging time points as inputs, and outputs the estimated integrated water stress value.

[0066] (Effect of the machine learning method S2 of the mechanical model) As described above, in the machine learning method S2 of the mechanical model, for one fruit in a fruit-bearing state on a tree, imaging is performed at a first time point and a second time point, and a set of two images with different imaging time points and the estimated integrated water stress value during the period between the first time point and the second time point estimated from the state of the tree on which the fruit is borne is used as teacher data. An acquisition step of acquiring a plurality of sets, and a learning step of learning to output an estimated integrated water stress value when two images, namely, the image taken at the first time point and the image taken at the second time point, are input for one fruit using the teacher data. A configuration including these steps is adopted.

[0067] According to the learning method S2 of the mechanical model configured as described above, the mechanical model can be trained to take an image of a fruit including temporal changes as input and output an estimated integrated water stress value.

[0068] (Configuration of the Brix Estimation Device 6) The configuration of the brix estimation device 6 according to the present embodiment will be described. The brix estimation device 6 is, for example, a device that estimates the brix of a fruit from the estimated integrated water stress values for a plurality of periods obtained by the integrated water stress value estimation device 1 or the integrated water stress value estimation device 1A. FIG. 8 is a block diagram showing a configuration example of the brix estimation device according to an embodiment of the present invention. In the example of FIG. 8, the brix estimation device 6 includes an integrated water stress value estimation unit 61, a totalization unit 62, a brix estimation unit 63, an output unit 64, and a mechanical model 65. Note that the configuration of the integrated water stress value estimation unit 61 is the same as the configuration of the integrated water stress value estimation device 1 or the integrated water stress value estimation device 1A described above. Alternatively, the brix estimation device 6 may include an integrated water stress value acquisition unit instead of the integrated water stress value estimation unit 61 (not shown). The integrated water stress value acquisition unit acquires, for example, the estimated integrated water stress value output by the integrated water stress value estimation device 1 or the integrated water stress value estimation device 1A described above.

[0069] (Totalization Unit 62) The totalization unit 62 totals, for example, the estimated integrated water stress values for a plurality of periods of one tree obtained by the integrated water stress value estimation unit 61.

[0070] (Brix Estimation Unit 63) The brix estimation unit 63 estimates, for example, the brix of the fruit that has set on the tree with reference to the totaled integrated water stress value. Further, the estimation process may be executed using the mechanical model 65 trained to output the estimated brix of the fruit when the integrated water stress values integrated over a plurality of periods are input.

[0071] (Output Unit 64) The output unit 64 outputs, for example, the estimated brix.

[0072] (Mechanical Model 65) The mechanical model 65 is a mechanical model that is trained to output the estimated sugar content of fruits when, for example, the number of days elapsed from when the flowers on the tree are in full bloom until an image of the fruit is captured, the estimated sugar content of the fruit in the image with the earliest imaging time among a plurality of images of the fruit, and the integrated water stress value integrated over a plurality of periods are input. As a specific example, in the case of fruit trees in the open field, the sugar content of the fruit that has set on the fruit tree reaches the minimum value a predetermined number of days after the flowers on the fruit tree are in full bloom, and thereafter, the amount of sugar increase in the fruit changes depending on the degree of water stress. For example, the mechanical model 65 may be a mechanical model trained to output the estimated sugar content of the fruit from the above-mentioned input values regarding such fruits.

[0073] (Effect by the Sugar Content Estimation Device 6) According to the sugar content estimation device 6 configured as described above, the sugar content of the fruit can be estimated from the estimated integrated water stress values for a plurality of periods obtained by the integrated water stress value estimation device 1 or the integrated water stress value estimation device 1A.

[0074] (Flow of the Sugar Content Estimation Method S3) The flow of the sugar content estimation method S3 executed by the sugar content estimation device 6 will be described. FIG. 9 is a flowchart showing an example of the flow of the sugar content estimation method S3 according to the present embodiment. Note that the processes in steps S31 to S33 are the same as the processes in steps S11 to S13 in the above-mentioned integrated water stress value estimation method S1. Alternatively, the sugar content estimation method S3 may include a step of obtaining the integrated water stress value instead of steps S31 to S33 (not shown).

[0075] (Step S34) In step S34, the totalizer 62 totals, for example, the estimated integrated water stress values for a plurality of periods of one tree obtained in steps S31 to S33.

[0076] (Step S35) In step S35, the sugar content estimation unit 63 estimates the sugar content of the fruits that have set on the tree, for example, by referring to the total integrated water stress value. Further, step S35 may be executed using a machine model 65 that has been learned to output the estimated sugar content of the fruit when, for example, the number of days elapsed from when the flowers on the tree were in full bloom until the fruit image was captured, the estimated value of the sugar content of the fruit in the image with the earliest imaging time among the plurality of fruit images, and the integrated water stress values integrated over a plurality of periods are input.

[0077] (Step S36) In step S36, the output unit 64 outputs, for example, the estimated sugar content.

[0078] (Effect by the sugar content estimation method S3) According to the sugar content estimation method S3 configured as described above, the sugar content of the fruit can be estimated from the estimated integrated water stress values for a plurality of periods obtained by the integrated water stress value estimation device 1 or the integrated water stress value estimation device 1A.

[0079] [Example of realization by software] The functions of the integrated water stress value estimation devices 1 and 1A, the machine models 2 and 2A, and the sugar content estimation device 6 (hereinafter referred to as "devices") are programs for causing a computer to function as the devices, and can be realized by programs for causing a computer to function as each control block (each unit included in the integrated water stress value estimation devices 1 and 1A, the machine models 2 and 2A, and the sugar content estimation device 6) of the devices.

[0080] In this case, the above device includes a computer having at least one control device (for example, a processor) and at least one storage device (for example, a memory) as hardware for executing the above program. By executing the above program by this control device and storage device, each function described in the above embodiments is realized.

[0081] The above program may be recorded on one or more computer-readable recording media, rather than temporarily. This recording medium may or may not be provided in the above device. In the latter case, the above program may be supplied to the above device via any wired or wireless transmission medium.

[0082] In addition, part or all of the functions of each of the above control blocks can also be realized by a logic circuit. For example, an integrated circuit in which a logic circuit functioning as each of the above control blocks is formed is also included in the scope of the present invention. In addition to this, for example, it is also possible to realize the functions of each of the above control blocks by a quantum computer.

[0083] 〔Summary〕 The integrated water stress value estimation method according to Aspect 1 of the present invention includes, for one fruit in a state of bearing fruit on a tree, an acquisition step of performing imaging at a first time point and a second time point, and acquiring two images with different imaging time points, and an input step of inputting the two acquired images into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image taken at the first time point and the image taken at the second time point, for one fruit are input, and an output step of outputting the estimated integrated water stress value output by the machine model.

[0084] According to the above configuration, the characteristics of the fruit can be extracted using a machine model, and the integrated water stress value can be estimated from the temporal change of the characteristics.

[0085] The integrated water stress value estimation method according to Aspect 2 of the present invention is, in the above Aspect 1, the machine model includes a first machine model that takes each of the two images as an input and outputs a matrix for each of the two images, where the number of rows and columns of the two matrices are the same, and a second machine model that takes the difference between the two output matrices as an input and outputs the estimated integrated water stress value.

[0086] According to the above configuration, by selectively using machine models for each processing step, the processing can be distributed to improve the processing efficiency of the entire system.

[0087] In the integrated water stress value estimation method according to Aspect 3 of the present invention, in the above Aspect 2, the first machine model includes a first A machine model that takes one of the two images, namely the image captured at the first time point and the image captured at the second time point, as an input and outputs a matrix having a predetermined number of rows and columns, and a first B machine model that takes the other of the two images, namely the image captured at the first time point and the image captured at the second time point, as an input and outputs a matrix having a predetermined number of rows and columns.

[0088] According to the above configuration, by processing two images with separate machine models, the processing can be distributed to further improve the processing efficiency of the entire system.

[0089] In the integrated water stress value estimation method according to Aspect 4 of the present invention, in any one of the above Aspects 1 to 3, the machine model is a machine model using a neural network.

[0090] According to the above configuration, the characteristics of the fruit can be extracted from the fruit image with higher accuracy.

[0091] The integrated water stress value estimation method according to Aspect 5 of the present invention further includes, in any one of the above Aspects 1 to 4, a step of inputting the output estimated integrated water stress value into an irrigation system.

[0092] According to the above configuration, the estimated integrated water stress value can be used to determine whether to irrigate the tree on which the fruit is set.

[0093] In the integrated water stress value estimation method according to Aspect 6 of the present invention, in any one of the above Aspects 1 to 5, the estimated integrated water stress value output by the machine model is the estimated integrated water stress value of the tree on which the fruit included in the two input images is set.

[0094] According to the above configuration, the integrated water stress value of a tree on which fruits have set can be estimated.

[0095] The integrated water stress value estimation device according to Aspect 7 of the present invention includes: an acquisition unit that captures an image of one fruit in a state of being set on a tree at a first time point and a second time point, and acquires two images with different imaging time points; an input unit that inputs the two acquired images into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, for one fruit are input; and an output unit that outputs the estimated integrated water stress value output by the machine model.

[0096] According to the above configuration, the same effect as in Aspect 1 is achieved.

[0097] The program according to Aspect 8 of the present invention causes at least one processor to execute: an acquisition step of capturing an image of one fruit in a state of being set on a tree at a first time point and a second time point, and acquiring two images with different imaging time points; an input step of inputting the two acquired images into a machine model learned to output an estimated integrated water stress value when two images, i.e., the image captured at the first time point and the image captured at the second time point, for one fruit are input; and an output step of outputting the estimated integrated water stress value output by the machine model.

[0098] According to the above configuration, the same effect as in Aspect 1 is achieved.

[0099] The method for learning a machine model according to Aspect 9 of the present invention includes, for one fruit in a state of bearing fruit on a tree, imaging at a first time point and a second time point, two images with different imaging time points, and an estimated integrated water stress value during a period between the first time point and the second time point estimated from the state of the tree on which the fruit has borne fruit. An acquisition step of acquiring a plurality of sets of the set data as teacher data; and a learning step of using the teacher data to learn to output an estimated integrated water stress value when two images, namely, the image captured at the first time point and the image captured at the second time point, are input for one fruit.

[0100] According to the above configuration, the machine model can be learned so as to take an image including the change over time of the fruit as an input and output the estimated integrated water stress value.

[0101] The method for learning a machine model according to Aspect 10 of the present invention is, in the above Aspect 9, the estimated integrated water stress value is evaluated by the following formula:

Equation

[0102] According to the above configuration, the estimated integrated water stress value can be evaluated using a mathematical formula.

[0103] The learning method of the mechanical model according to Aspect 11 of the present invention is, in the above Aspect 10, the predetermined value is -0.4 MPa.

[0104] According to the above configuration, after setting a predetermined value of water potential, the estimated integrated water stress value can be evaluated.

[0105] The learned mechanical model according to Aspect 12 of the present invention captures an image of one fruit in a state of bearing fruit on a tree at a first time point and a second time point, takes two images with different imaging time points as input, and outputs an estimated integrated water stress value.

[0106] According to the above configuration, the integrated water stress value can be estimated from an image including the temporal change of the fruit.

[0107] The computer-readable non-transitory recording medium according to Aspect 13 of the present invention records the program according to Aspect 8 of the present invention.

[0108] According to the above configuration, the same effect as that of Aspect 8 is achieved.

[0109] The method for estimating the sugar content of a fruit according to Aspect 14 of the present invention includes a total step of totaling the estimated integrated water stress values in a plurality of periods of one tree obtained by the integrated water stress value estimation method according to any one of the above Aspects 1 to 6, an estimation step of estimating the sugar content of the fruit bearing on the tree with reference to the totaled integrated water stress value, and an output step of outputting the estimated sugar content.

[0110] According to the above configuration, the sugar content of the fruit can be estimated from the estimated integrated water stress values in a plurality of periods.

[0111] The method for estimating the sugar content of fruits according to Embodiment 15 of the present invention is, in the above Embodiment 14, when the number of days elapsed from when the flowers on the tree are in full bloom until the image of the fruit is captured, the estimated value of the sugar content of the fruit in the image with the earliest imaging time among the plurality of images of the fruit, and the integrated water stress value integrated over a plurality of periods are input, it is executed using a machine model learned to output the estimated sugar content of the fruit.

[0112] According to the above configuration, the sugar content of the fruit can be estimated using a machine model from the estimated integrated water stress values for a plurality of periods.

[0113] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope shown in the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

Explanation of Reference Numerals

[0114] 1, 1A Integrated water stress value estimation device 2, 2A, 14, 65 Machine model 3 Output device 4 Irrigation system 5 Network 6 Sugar content estimation device 11 Acquisition unit 12 Input unit 13, 64 Output unit 15 Communication unit 21 First machine model 22 Second machine model 61 Integrated water stress value estimation unit 62 Total unit 63 Sugar content estimation unit 100 Database

Claims

1. an acquisition step of acquiring two images of a fruit at different times by capturing images of the fruit at a first time point and a second time point, the two images being captured at different times; an input step of inputting the two acquired images into a machine model that has been trained to output an estimated accumulated water stress value when two images of one fruit, an image taken at the first time point and an image taken at the second time point, are input; an output step of outputting the estimated integrated water stress value output by the mechanical model; A method for estimating an integrated water stress value,

2. The accumulated water stress value estimation method of claim 1, wherein the mechanical model comprises: a first mechanical model that takes each of the two images as input and outputs a matrix relating to each of the two images, the matrix having the same number of rows and columns, and a second mechanical model that takes the difference between the two output matrices as input and outputs the estimated accumulated water stress value.

3. The method for estimating an accumulated water stress value as described in claim 2, wherein the first mechanical model includes a firstA mechanical model that takes one of two images, an image taken at the first time point and an image taken at the second time point, as input and outputs a matrix having a predetermined number of rows and columns, and a firstB mechanical model that takes the other of the two images, an image taken at the first time point and an image taken at the second time point, as input and outputs a matrix having a predetermined number of rows and columns.

4. The method for estimating an integrated water stress value according to claim 1 , wherein the machine model is a machine model using a neural network.

5. The method for estimating an integrated water stress value according to claim 1 , further comprising the step of inputting the output estimated integrated water stress value into an irrigation system.

6. An accumulated water stress value estimation method according to any one of claims 1 to 3, wherein the estimated accumulated water stress value output by the machine model is an estimated accumulated water stress value of a tree bearing the fruit contained in the two input images.

7. an acquisition unit that captures images of a single fruit on a tree at a first time point and a second time point to acquire two images captured at different times; an input unit that inputs the two acquired images into a machine model that has been trained to output an estimated accumulated water stress value when two images of one fruit, an image taken at the first time point and an image taken at the second time point, are input; an output unit that outputs the estimated integrated water stress value output by the mechanical model; An integrated water stress value estimating device comprising:

8. At least one processor, an acquisition step of acquiring two images of a fruit at different times by capturing images of the fruit at a first time point and a second time point, the two images being captured at different times; an input step of inputting the two acquired images into a machine model that has been trained to output an estimated accumulated water stress value when two images of one fruit, an image taken at the first time point and an image taken at the second time point, are input; an output step of outputting the estimated integrated water stress value output by the mechanical model; A program to execute.

9. an acquisition step of capturing images of a single fruit on a tree at a first time point and a second time point, and acquiring, as teacher data, a plurality of sets of data including two images captured at different times and an estimated integrated water stress value for the period between the first time point and the second time point, which is estimated from the state of the tree on which the fruit is borne; a learning step of learning to output an estimated integrated water stress value when two images of one fruit, an image taken at the first time point and an image taken at the second time point, are inputted using the teacher data; How to train machine models, including:

10. The estimated integrated water stress value is evaluated by the following formula: [0010] Where: SΨ ｘ : Accumulated water stress value from the time the first image was taken to the time the second image was taken (target period) Psi ｉ : Measured water stress value of the target tree leaf at the time of capturing the first image Psi ｉ＋１ : Measured water stress value of the target tree leaf at the time the second image was taken c: Ψ of the target tree during the target period ｍａｘ (Water potential) (July-September: Predetermined value, October onwards: Average value of humid area or low stress area) n: Number of days in the target period The method for learning a machine model according to claim 9 , wherein

11. The machine model learning method according to claim 10, wherein the predetermined value is −0.4 MPa.

12. A trained machine model in which images of a single fruit on a tree are taken at a first and a second time point, and the two images taken at different times are used as input, and the estimated accumulated water stress value is output.

13. A computer-readable non-transitory recording medium having the program according to claim 8 recorded thereon.

14. a summing step of summing the estimated integrated water stress values ​​for one tree over a plurality of time periods, the estimated integrated water stress values ​​being obtained by the integrated water stress value estimation method according to any one of claims 1 to 3; an estimation step of estimating the sugar content of a fruit borne on the tree by referring to the summed integrated water stress value; An output step of outputting the estimated sugar content; A method for estimating sugar content of fruit, comprising:

15. The method for estimating sugar content of a fruit as described in claim 14, wherein the estimation step is performed using a machine model that has been trained to output an estimated sugar content of the fruit when the number of days that have elapsed since the flowers on the tree reached full bloom until the image of the fruit was captured, an estimated sugar content of the fruit in the earliest image among the multiple images of the fruit, and an integrated water stress value accumulated over multiple periods are input.

Images