JP7638491B2 - Information processing device, information processing system, information processing method, and computer program - Google Patents

Description

The technology disclosed in this specification relates to an information processing device that estimates the size of a plant under cultivation.

In recent years, various technologies have been proposed to improve the efficiency and optimization of plant cultivation (leafy vegetables, fruit trees, ornamental plants, etc.). For example, a technology is known in which, in fruit cultivation carried out by dedicated farmers, the size of the fruit and the integrated value of the leaf area are calculated from images of the fruit trees, a growth curve is determined that approximates the growth state of the fruit based on the integrated value of the fruit size and leaf area, the growth curve is used to evaluate the growth state of the cultivated fruit, and the harvest date of the fruit is predicted based on the evaluation results (see, for example, Patent Document 1).

JP 2019-187259 A

In the conventional technology described above, the size of each fruit in the image is measured manually, which means that it takes a lot of time and effort to determine the size of the fruit. This problem is not limited to determining the size of fruit to predict the harvest date, but is a common problem when determining (estimating) the size of plants under cultivation.

This specification discloses a technology that can solve the above-mentioned problems.

The technology disclosed in this specification can be realized, for example, in the following forms:

(1) The information processing device disclosed in this specification includes a target data acquisition unit and an average size estimation unit. The target data acquisition unit acquires target image data, which is image data generated by photographing a cultivation area in which a plurality of target plants, which are specific types of plants, are cultivated. The average size estimation unit estimates an average size, which is the average value of the sizes of the target plants cultivated in the cultivation area, using image data including color information of the target image data and a model created by a predetermined machine learning. In this way, the information processing device acquires target image data generated by photographing a cultivation area, and can estimate the average size of the target plants using image data including color information of the target image data and a model created by machine learning. Therefore, according to the information processing device, the average size of the target plants can be estimated simply by acquiring the target image data, and the effort and time required for estimating the average size of the target plants can be reduced.

(2) The information processing device may further include a training data acquisition unit that acquires training data that associates a plurality of training image data, which is image data generated by photographing a cultivation area in which a plurality of target plants are cultivated, at a plurality of dates and times, with average size data that indicates the average size of the target plants in the cultivation area represented by the training image data, and a model acquisition unit that acquires the model by the machine learning using the training data. According to this information processing device, it is possible to acquire a model for estimating the average size of the target plants without using another device, and to estimate the average size of the target plants using the model, thereby further reducing the effort and time required to estimate the average size of the target plants.

(3) In the above information processing device, the training image data and the target image data may be configured to include overhead image data, which is image data generated by photographing the cultivation area with a photographing device in which the angle between the optical axis of the lens and the horizontal plane is 20 degrees or more. According to this information processing device, the average size can be estimated using an image in which the cultivation area is represented in a relatively wide image area, and the accuracy of estimating the average size of the target plant can be improved.

(4) In the above information processing device, the training image data and the target image data include, in addition to the overhead image data, horizontal image data, which is image data generated by photographing the cultivation area with a photographing device in which the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees, and the model acquisition unit acquires the first model by machine learning using the training data for the overhead image data and acquires the second model by machine learning using the training data for the horizontal image data, and the average size estimation unit estimates the average size using the first model for the target image data that is the overhead image data and estimates the average size using the second model for the target image data that is the horizontal image data. According to this information processing device, the average size can be estimated using a model optimized according to the photographing angle of the image, and the accuracy of estimating the average size of the target plant can be further improved.

(5) In the above information processing device, the training image data and the target image data may be image data generated by photographing the cultivation area during a preset daytime period. According to this information processing device, the average size can be estimated using an image in which the color of the target plant cultivated in the cultivation area is relatively faithfully represented, and the accuracy of estimating the average size of the target plant can be improved.

(6) In the above information processing device, the training image data and the target image data may be image data generated by photographing an area including a plurality of the cultivation sections and cropping each of the cultivation sections. According to this information processing device, even if the growth conditions of the target plants differ among the cultivation sections, the average size of the target plants cultivated in each cultivation section can be estimated with high accuracy.

(7) In the above information processing device, the trimming may be configured to be performed so that the trimmed image for one of the cultivation sections does not include pixels representing the target plants being cultivated in the other cultivation sections. According to this information processing device, it is possible to eliminate the influence of pixels representing target plants in the other cultivation sections, and to estimate the average size of the target plants being cultivated in each cultivation section with even higher accuracy.

(8) The information processing device may further include a harvest date and time prediction unit that predicts a harvest date and time of the target plant based on the average size of the target plant estimated from the target image data. According to this information processing device, it is possible to predict the harvest date and time of the target plant simply by acquiring the target image data, and it is possible to effectively realize the efficiency and optimization of plant cultivation.

(9) In the above information processing device, the target data acquisition unit may acquire a plurality of sets of target image data generated by photographing the cultivation area at a plurality of dates and times, the average size estimation unit may estimate an average size of the target plant for the plurality of sets of target image data, and the harvest date and time prediction unit may predict a harvest date and time of the target plant for one set of target image data based on the average size of the target plant estimated for the one set of target image data and the average size of the target plant estimated for each set of target image data photographed before the photographing date and time of the one set of target image data. According to this information processing device, the harvest date and time of the target plant can be predicted with high accuracy, and the efficiency and optimization of plant cultivation can be more effectively achieved.

(10) In the above information processing device, the target plant may be a leafy vegetable, and the average size may be the average plant height of the target plant. According to this information processing device, it is possible to reduce the effort and time required to estimate the average plant height of a leafy vegetable.

(11) The information processing system disclosed in this specification includes the information processing device and a terminal device capable of communicating with the information processing device via a communication network. The information processing device further includes an information transmission unit that transmits information indicating the estimated average size of the target plant to the terminal device via the communication network. According to this information processing system, the information transmission unit of the information processing device transmits information indicating the estimated average size of the target plant to the terminal device, so that a person holding the terminal device can grasp the average size of the target plant even if he or she is in a location other than the field, and the efficiency and optimization of plant cultivation can be more effectively achieved.

(12) The above information processing system may further include a photographing device that is installed in the field where the target plant is cultivated and is capable of communicating with the information processing device via a communication network, and the target data acquisition unit may be configured to acquire the target image data generated by the photographing device photographing the cultivation area via the communication network. According to this information processing system, the photographer (e.g., a farm worker) can acquire the target image data without going to the field, and the average size of the target plant can be estimated using the acquired target image data, thereby making it possible to extremely effectively realize efficiency and optimization of plant cultivation.

The technology disclosed in this specification can be realized in various forms, such as an information processing device, an information processing system, an information processing method, a computer program that realizes these methods, and a non-transitory recording medium on which the computer program is recorded.

FIG. 1 is an explanatory diagram showing a schematic configuration of a plant cultivation management system 10 according to an embodiment of the present invention. A block diagram showing a schematic configuration of a plant cultivation management server 100. A flowchart showing the contents of an average size estimation model creation process in this embodiment. FIG. 1 is an explanatory diagram conceptually illustrating an example of training data TD. FIG. 1 is an explanatory diagram conceptually illustrating an example of training data TD. FIG. 6 is an explanatory diagram showing the conditions for trimming image data I. FIG. 1 is an explanatory diagram conceptually illustrating an example of an average size estimation model MO. A flowchart showing the contents of a plant cultivation support process according to the present embodiment. FIG. 2 is an explanatory diagram showing an example of a screen displayed on the display unit 210 of the terminal device 200 during the plant cultivation support process in the present embodiment. FIG. 1 is an explanatory diagram showing an example of a method for predicting a harvest date and time Te based on an average size AS in this embodiment. FIG. 2 is an explanatory diagram showing an example of a screen displayed on the display unit 210 of the terminal device 200 during the plant cultivation support process in the present embodiment. FIG. 13 is an explanatory diagram showing the prediction accuracy of the average size AS for each average size estimation model MO. FIG. 13 is an explanatory diagram showing the prediction accuracy of the average size AS for each trimming condition.

A. Embodiments:
A-1. Configuration of the plant cultivation management system 10:
FIG. 1 is an explanatory diagram showing a schematic configuration of a plant cultivation management system 10 in this embodiment. The plant cultivation management system 10 is a system for managing the cultivation of plants in a farm field. More specifically, the plant cultivation management system 10 is a system that uses image data I generated by photographing a cultivation section PL in which a plurality of specific types of plants (hereinafter referred to as "target plants OP") are cultivated, to estimate an average size, which is an average value of the sizes of each target plant OP cultivated in the cultivation section PL, and aims to improve the efficiency and optimization of plant cultivation. In this embodiment, the farm field is a semi-open type (sunlight utilization type) plant factory PF. Here, the semi-open type plant factory PF is a farm field surrounded by a material that does not block sunlight, such as a vinyl greenhouse. However, the plant cultivation management system 10 can also be applied to other types of farm fields (for example, other types of plant factories such as closed type and open type, fields for outdoor cultivation, etc.). Although only one plant factory PF is shown in FIG. 1, the plant cultivation management system 10 can also be applied to the management of plant cultivation in each of multiple plant factories PF. In this embodiment, the target plant OP is spinach, which is a leafy vegetable, and the average size is the average plant height. However, the plant cultivation management system 10 can also be applied to estimate other average sizes (for example, average fruit diameters) for other plants (leafy vegetables other than spinach, fruit trees, ornamental plants, etc.). The plant cultivation management system 10 is an example of an information processing system in the claims.

The plant cultivation management system 10 includes a plant cultivation management server 100, a terminal device 200, and a photography device 300 installed in the plant factory PF. Each device constituting the plant cultivation management system 10 is connected to each other so as to be able to communicate with each other via a communication network NET.

The photographing device 300 is a device that is installed in the plant factory PF and generates image data I (e.g., RGB data) by photographing a cultivation section PL in which multiple target plants OP are cultivated. In this embodiment, the photographing device 300 is a web camera that takes photographs periodically (e.g., every hour) or continuously of a preset range in the plant factory PF, specifically, a range covering one or multiple cultivation sections PL, as a subject. The photographing device 300 also has a communication interface (not shown), and transmits image data I generated by photographing each time it takes photographs to the plant cultivation management server 100 via the communication network NET. Note that, when there are multiple plant factories PF that are managed by the plant cultivation management system 10, the photographing device 300 is installed in each plant factory PF.

In addition, in this embodiment, two types of imaging devices 300 are used: an overhead imaging device 300A and a horizontal imaging device 300B. The overhead imaging device 300A is an imaging device 300 installed at a position where the angle between the optical axis of the lens and the horizontal plane is 20 degrees or more, and generates image data I (hereinafter referred to as "overhead image data Io") that captures the cultivation section PL from an overhead view. On the other hand, the horizontal imaging device 300B is an imaging device 300 installed at a position where the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees, and generates image data I (hereinafter referred to as "horizontal image data Ih") that captures the cultivation section PL from a direction close to horizontal. Both the overhead imaging device 300A and the horizontal imaging device 300B may be installed in each plant factory PF, or either one of the overhead imaging device 300A or the horizontal imaging device 300B may be installed.

The terminal device 200 is, for example, a smartphone or a tablet terminal. The terminal device 200 has a display unit 210 configured, for example, by a liquid crystal display or the like (see Figures 9 and 11). In addition, an application program for plant cultivation management is installed in the terminal device 200. As will be described in more detail below, a user (farmer) of the terminal device 200 can use the application program to ascertain the average size of the target plants OP being cultivated in the plant factory PF and the predicted harvest date and time.

The plant cultivation management server 100 is a server device for managing the cultivation of plants in the plant factory PF. FIG. 2 is a block diagram showing a schematic configuration of the plant cultivation management server 100. The plant cultivation management server 100 includes a control unit 110, a memory unit 120, a display unit 130, an operation input unit 140, and an interface unit 150. These units are connected to each other via a bus 190 so as to be able to communicate with each other. The plant cultivation management server 100 is an example of an information processing device within the scope of the claims.

The display unit 130 of the plant cultivation management server 100 is, for example, a liquid crystal display, and displays various images and information. The operation input unit 140 is, for example, a keyboard, mouse, buttons, microphone, and the like, and accepts operations and instructions from the administrator. The display unit 130 may be equipped with a touch panel so as to function as the operation input unit 140. The interface unit 150 is, for example, a LAN interface, a USB interface, and the like, and communicates with other devices via wired or wireless connection.

The storage unit 120 of the plant cultivation management server 100 is composed of, for example, a ROM, a RAM, a hard disk drive (HDD), etc., and is used to store various programs and data, and as a working area when executing various programs, and as a temporary storage area for data. For example, the storage unit 120 stores a plant cultivation management program CP, which is a computer program for executing the average size estimation model creation process and the plant cultivation support process described below. The plant cultivation management program CP is provided in a state stored in a computer-readable recording medium (not shown), such as a CD-ROM, DVD-ROM, or USB memory, and is stored in the storage unit 120 by installing it in the plant cultivation management server 100.

In addition, the storage unit 120 of the plant cultivation management server 100 stores training data TD, average size estimation model MO, target image data I(o), estimated average size data GD, and predicted harvest date and time data PD during the average size estimation model creation process and plant cultivation support process described below. The contents of this information will be explained in conjunction with the explanation of the average size estimation model creation process and plant cultivation support process described below.

The control unit 110 of the plant cultivation management server 100 is configured, for example, by a CPU, and controls the operation of the plant cultivation management server 100 by executing a computer program read from the storage unit 120. For example, the control unit 110 functions as a plant cultivation management unit 111 that executes an average size estimation model creation process and a plant cultivation support process, described below, by reading and executing a plant cultivation management program CP from the storage unit 120. The plant cultivation management unit 111 includes a training data acquisition unit 112, a model acquisition unit 114, a target data acquisition unit 115, an average size estimation unit 116, a harvest date and time prediction unit 117, and an information transmission unit 118. The functions of each of these units will be described in conjunction with the explanation of the average size estimation model creation process and the plant cultivation support process, described below.

A-2. Average size estimation model creation process:
Next, an average size estimation model creation process executed by the plant cultivation management server 100 of this embodiment will be described. FIG. 3 is a flowchart showing the contents of the average size estimation model creation process in this embodiment. The average size estimation model creation process is a process for creating an average size estimation model MO for estimating an average size (average plant height in this embodiment), which is an average value of the sizes of each target plant OP (spinach in this embodiment) cultivated in the cultivation section PL, from image data I generated by photographing the cultivation section PL in which a plurality of target plants OP are cultivated, using a predetermined machine learning. The average size estimation model creation process is started in response to an administrator operating the operation input unit 140 of the plant cultivation management server 100 to input a start instruction.

First, the training data acquisition unit 112 (FIG. 2) of the plant cultivation management server 100 acquires training data TD (S110). The training data TD is data in which multiple training image data I(t) generated by photographing the cultivation area PL with the photographing device 300 in the plant factory PF are associated with average size data indicating the average size of the target plants OP in the cultivation area PL represented by each training image data I(t). The acquired training data TD is stored in the memory unit 120 of the plant cultivation management server 100.

4 and 5 are explanatory diagrams conceptually illustrating an example of training data TD. FIG. 4 shows an example of training data TD using overhead image data Io, and FIG. 5 shows an example of training data TD using horizontal image data Ih. As described above, the photographing devices 300 (overhead photographing device 300A and horizontal photographing device 300B) installed in the plant factory PF take photographs periodically, and each time photographing is performed, the image data I (overhead image data Io and horizontal image data Ih) generated by the photographing is transmitted to the plant cultivation management server 100 via the communication network NET. The training data acquisition unit 112 of the plant cultivation management server 100 acquires the image data I transmitted from the photographing device 300 via the interface unit 150 and stores it in the memory unit 120 as training image data I(t).

As described above, in this embodiment, the photographing device 300 photographs an area covering one or more cultivation sections PL as a subject. Therefore, the training data acquisition unit 112 performs trimming on the image data I transmitted from the photographing device 300 for each cultivation section PL, and stores the trimmed image data I in the memory unit 120 as training image data I(t). Figure 6 is an explanatory diagram showing the conditions (hereinafter referred to as "trimming conditions") when trimming the image data I. In this embodiment, three trimming conditions (trimming conditions 1 to 3) are set. Column A of FIG. 6 shows an example of a trimming frame TF corresponding to each cultivation section PL (PL1-3) when the overhead image data Io is trimmed according to trimming condition 1, column B of FIG. 6 shows an example of a trimming frame TF corresponding to each cultivation section PL when the overhead image data Io is trimmed according to trimming condition 2, and column C of FIG. 6 shows an example of a trimming frame TF corresponding to each cultivation section PL when the overhead image data Io is trimmed according to trimming condition 3.

Cropping condition 1 (see column A in Figure 6) is a condition for performing cropping such that, for each cultivation section PL, among the pixels depicting the target plants OP being cultivated in the cultivation section PL, the pixels closest to the top, bottom, left and right edges of the image become the top, bottom, left and right edges of the cropped image. In other words, cropping according to cropping condition 1 results in image data I that depicts the entirety of multiple target plants OP being cultivated for each cultivation section PL.

Moreover, trimming condition 2 (see column B in Figure 6) is a condition for trimming based on trimming condition 1 so as to include as many pixels as possible and not include pixels that show target plants OP cultivated in other cultivation sections PL. In other words, by trimming according to trimming condition 2, image data I is obtained for each cultivation section PL that does not show target plants OP cultivated in other cultivation sections PL.

Cropping condition 3 (see column C in Figure 6) is a cropping condition based on cropping condition 2, in which both ends of the long side of the image are cut so that the vertical and horizontal sizes of the image are the same. In other words, cropping according to cropping condition 3 results in image data I that is free from image distortion when the image data I is converted to data with an aspect ratio of 1:1 for learning purposes.

As shown in FIG. 4 and FIG. 5, hereinafter, image data I (overhead image data Io and horizontal image data Ih) photographed and generated at date and time Ti will be represented as image data Ii (overhead image data Ioi and horizontal image data Ihi). Image data I photographed at planting date and time Ts is image data Is (overhead image data Ios and horizontal image data Ihs), and image data I photographed at harvest date and time Te is image data Ie (overhead image data Ioe and horizontal image data Ihe). The training data acquisition unit 112 identifies the image data Is photographed at planting date and time Ts and the image data Ie photographed at harvest date and time Te from among the multiple image data stored in the storage unit 120 by a known image recognition process. In addition, the training data acquisition unit 112 selects image data I whose photographing date and time is within the period from planting date and time Ts to harvest date and time Te from among the multiple image data I stored in the storage unit 120. However, in this embodiment, the target is a semi-open plant factory PF where cultivation is performed using sunlight, and only image data I generated by photographing during a preset daytime period (e.g., 7:00 to 17:00) is used; image data I photographed during other times is excluded because there is a risk that the plant's shape will not be clearly visible in the image data I. In this way, the training data acquisition unit 112 selects multiple image data Ii (Is, I1, I2, ..., Ie) photographed and generated at multiple dates and times Ti (Ts, T1, T2, ..., Te) during the period from the planting date and time Ts to the harvest date and time Te as training image data I(t).

The training data acquisition unit 112 also acquires the value of the average size AS (average plant height) of the target plant OP for each of the multiple image data I stored in the storage unit 120. The average size AS of the target plant OP can be measured, for example, by referring to an image of a ruler that appears together with the target plant OP in the image data I. The average size AS of the target plant OP may also be acquired by referring to a document showing the actual measurement value measured by a farm worker on-site at the plant factory PF and recorded together with the date and time. The training data acquisition unit 112 creates training data TD as labeled data by associating the acquired value of the average size AS with the training image data I(t). For example, as shown in FIG. 4, if the average size AS of the target plant OP is 7 cm for the overhead image data Iom captured at a certain date and time Tm, data in which the value of 7 cm is associated with the overhead image data Iom as the average size ASm is created as training data TD. The same applies to other image data Ii.

The training data acquisition unit 112 may acquire training data TD for one crop of the target plant OP, or may acquire training data TD for multiple crops of the target plant OP in order to improve the accuracy of the average size estimation model MO. In addition, when multiple image capture devices 300 are installed in one plant factory PF, the training data acquisition unit 112 may acquire training data TD based on image data acquired from the multiple image capture devices 300. In addition, the training data acquisition unit 112 may acquire training data TD based on image data acquired from the image capture devices 300 installed in each of the multiple plant factories PF. In addition, the training image data I(t) may be generated by performing data augmentation (data expansion) on the image data acquired from the image capture devices 300 by rotating, shifting, adjusting brightness, etc.

Next, the model acquisition unit 114 (FIG. 2) of the plant cultivation management server 100 creates an average size estimation model MO by a predetermined machine learning using the training data TD (S120 in FIG. 3). FIG. 7 is an explanatory diagram conceptually showing an example of the average size estimation model MO. As shown in FIG. 7, the average size estimation model MO is a model for estimating the average size AS (i.e., the average size AS at the shooting date and time) of the target plant OP cultivated in the cultivation section PL represented by the image data I generated by photographing the cultivation section PL of the target plant OP from the image data I including color information (e.g., RGB values). In this embodiment, as the average size estimation model MO, an overhead average size estimation model MOo for estimating the average size AS of the target plant OP from the overhead image data Io and a horizontal average size estimation model MOh for estimating the average size AS of the target plant OP from the horizontal image data Ih are created.

The model acquisition unit 114 uses the training image data I(t) included in the training data TD for each of the bird's-eye view and the horizontal view as input data, the average size AS associated with each training image data I(t) as the objective variable, and performs known machine learning using at least one of the root mean square error (RMSE), the mean absolute error (MAE), and the coefficient of determination (R ² ) as the evaluation index to create an average size estimation model MO (an average size estimation model MOo for bird's-eye view and an average size estimation model MOh for horizontal view). The created average size estimation model MO is stored in the storage unit 120 of the plant cultivation management server 100. In addition, various known machine learning algorithms such as multiple regression analysis, Random Forest, XGBoost, and convolutional neural network (CNN) can be used to create the average size estimation model MO. The average size estimation model MO is an example of a model in the claims, the overhead average size estimation model MOo is an example of a first model in the claims, and the horizontal average size estimation model MOh is an example of a second model in the claims. Through the above steps, the growth rate estimation model creation process for creating the average size estimation model MO is completed.

In this embodiment, the overhead average size estimation model MOo is created using training data TD for the overhead image data Io, and the horizontal average size estimation model MOh is created using training data TD for the horizontal image data Ih, but it is also possible to create a single common average size estimation model MO using training data TD that is a mixture of overhead image data Io and horizontal image data Ih without distinguishing between overhead and horizontal use.

A-3. Plant cultivation support treatment:
Next, a plant cultivation support process executed by the plant cultivation management server 100 of this embodiment will be described. Fig. 8 is a flowchart showing the contents of the plant cultivation support process in this embodiment. Figs. 9 and 11 are explanatory diagrams showing an example of a screen displayed on the display unit 210 of the terminal device 200 during the plant cultivation support process in this embodiment, and Fig. 10 is an explanatory diagram showing an example of a method for predicting a harvest date and time Te based on the average size AS in this embodiment.

The plant cultivation support process is a process for supporting the cultivation of a target plant OP by a farm worker. More specifically, the plant cultivation support process is a process for estimating the average size AS (i.e., the average size AS at the photographing date and time of the target image data I(o)) of the target plant OP cultivated in the cultivation section PL represented by the target image data I(o) using image data including color information of the target image data I(o), which is image data I generated by photographing the cultivation section PL in which multiple target plants OP are cultivated, and the above-mentioned average size estimation model MO. Furthermore, the plant cultivation support process is a process for predicting the harvest date and time Te of the target plant OP based on the estimated average size AS, and displaying information indicating the average size AS and the harvest date and time Te on the farm worker's terminal device 200. The plant cultivation support process is started in response to input of a start instruction by the farm worker operating the terminal device 200 or the administrator of the plant cultivation management server 100 operating the operation input unit 140.

In the plant cultivation support process, first, the target data acquisition unit 115 (FIG. 2) of the plant cultivation management server 100 acquires the target image data I(o) (S210). As described above, the photographing device 300 installed in the plant factory PF periodically takes photographs, and each time photographs are taken, the image data I generated by the photographing is transmitted to the plant cultivation management server 100 via the communication network NET. The target data acquisition unit 115 of the plant cultivation management server 100 acquires the image data I transmitted from the photographing device 300 via the interface unit 150 and stores it in the storage unit 120 as the target image data I(o). As with the above-mentioned training data TD, the target data acquisition unit 115 trims the image data I transmitted from the photographing device 300 for each cultivation section PL, and stores the trimmed image data I in the storage unit 120 as the target image data I(o).

Next, the average size estimation unit 116 (FIG. 2) of the plant cultivation management server 100 estimates the average size AS of the target plant OP cultivated in the cultivation section PL represented by the target image data I(o) (i.e., the average size AS at the photographing date and time of the target image data I(o)) using the target image data I(o) acquired in the immediately preceding S210 and the above-mentioned average size estimation model MO (S220). As shown in FIG. 7, the average size estimation unit 116 inputs image data including color information of the target image data I(o) into the average size estimation model MO, and estimates the average size AS by acquiring the average size AS output from the average size estimation model MO. For example, in the example shown in FIG. 7, the average size AS of the target plant OP is estimated to be "10 cm". The average size estimation unit 116 generates estimated average size data SD, which is information indicating the estimated average size AS, and stores it in the storage unit 120 of the plant cultivation management server 100.

In this embodiment, two average size estimation models MO are created: an average size estimation model MOo for bird's-eye view and an average size estimation model MOh for horizontal view. The average size estimation unit 116 determines whether the target image data I(o) is bird's-eye view image data Io or horizontal image data Ih, and estimates the average size AS using the average size estimation model MOo for bird's-eye view if it is bird's-eye view image data Io, and estimates the average size AS using the average size estimation model MOh if it is horizontal image data Ih. The determination of whether the target image data I(o) is bird's-eye view image data Io or horizontal image data Ih can be realized, for example, by analyzing the image contents of the target image data I(o) or by referring to the attached information of the target image data I(o). In addition, the determination may be realized by creating a model for the determination by machine learning and performing classification using the model.

Next, the plant cultivation management unit 111 (FIG. 2) of the plant cultivation management server 100 judges whether the number N of the target image data I(o) for which the average size AS was estimated is equal to or greater than a predetermined threshold Th (e.g., 30) (S230). If it is judged in the judgment of S230 that the number N of the target image data I(o) for which the average size AS was estimated is less than the threshold Th (S230: NO), the prediction of the harvest date and time Te (described later) is not executed because it is difficult to formulate. In this case, the information transmission unit 118 (FIG. 2) of the plant cultivation management server 100 transmits the target image data I(o) acquired in the immediately preceding S210 and the estimated average size data SD indicating the average size AS estimated in the immediately preceding S220 to the farmer's terminal device 200 via the interface unit 150 and the communication network NET (S250). As a result, as shown in FIG. 9, the display unit 210 of the terminal device 200 displays a current (more precisely, the time when the target image data I(o) was taken) photo and estimated average size AS of the target plant OP (spinach in this embodiment) being cultivated in the plant factory PF. This allows the farm worker to know the current average size AS of the target plant OP without having to go to the plant factory PF. Note that, as described above, if it is determined in the judgment of S230 that the number N of target image data I(o) for which the average size AS was estimated is less than the threshold value Th, the prediction of the harvest date and time Te (S240) has not been performed, and therefore the symbol "-" indicating unpredictable is displayed in the column for the predicted harvest date and time Te.

After the process of S250, the plant cultivation management unit 111 (FIG. 2) of the plant cultivation management server 100 judges whether the cultivation has ended (whether the plant has been harvested) (S270). If it is judged in the judgment of S270 that the cultivation has not yet ended (S270: NO), the process from S210 onwards is repeatedly executed.

The process from S210 onwards is repeatedly executed, and if it is determined in S230 that the number N of the target image data I(o) for which the average size AS was estimated is equal to or greater than the threshold value Th (S230: YES), the harvest date and time prediction unit 117 (FIG. 2) of the plant cultivation management server 100 predicts the harvest date and time Te of the target plant based on the average size AS estimated for each target image data I(o) (S240). As shown in FIG. 10, in this embodiment, a regression equation is created by simple regression analysis using the average size ASi as the objective variable and the elapsed time ΔTi (=Ti-Ts) from the planting date and time Ts as the explanatory variable, and the harvest date and time Te is predicted by back-calculating the elapsed time ΔTi at which the average size AS becomes a predetermined harvest average size ASe (for example, ASe = 20 cm) in the regression equation. For example, in FIG. 10, when the estimation of the average size AS for the target image data I(o) captured at the date and time Tk is completed, (k+1) points (ΔTi, ASi) are plotted, which indicate the data for which the average size AS has been estimated so far. The harvest date and time prediction unit 117 uses these (k+1) pieces of data to derive the simple regression equation RL: ASi = a × ΔTi + b. The harvest date and time prediction unit 117 then calculates the value of ΔTi (= ΔTe) where ASi = ASe (for example, 20 cm) in this simple regression equation RL, and calculates the date and time (= Ts + ΔTe) at which this elapsed time ΔTe has elapsed from the planting date and time Ts as the predicted harvest date and time Te. The harvest date and time prediction unit 117 generates predicted harvest date and time data PD, which is information indicating the predicted harvest date and time Te, and stores it in the memory unit 120 of the plant cultivation management server 100.

When the harvest date and time Te is predicted (S240), the information transmission unit 118 (FIG. 2) of the plant cultivation management server 100 transmits the target image data I(o) acquired in the immediately preceding S210, the estimated average size data SD indicating the average size AS estimated in the immediately preceding S220, and the predicted harvest date and time data PD indicating the harvest date and time Te predicted in the immediately preceding S240 to the farmer's terminal device 200 via the interface unit 150 and the communication network NET (S260). As a result, as shown in FIG. 11, the display unit 210 of the terminal device 200 displays a photo of the target plant OP (spinach in this embodiment) cultivated in the plant factory PF at the current time (more precisely, at the time of taking the target image data I(o)), the estimated average size AS, and the predicted harvest date and time Te. This allows the farmer to grasp the current average size AS of the target plant and the predicted harvest date and time Te without going to the plant factory PF.

After the process of S260, the plant cultivation management unit 111 (FIG. 2) of the plant cultivation management server 100 judges whether cultivation has ended (whether the plant has been harvested) (S270). If it is judged in the judgment of S270 that cultivation has not yet ended (S270: NO), the process from S210 onwards is repeatedly executed. The above-mentioned process is repeatedly executed, and if it is judged in the judgment of S270 that cultivation has ended (S270: YES), the plant cultivation support process is terminated.

A-4. Advantages of this embodiment:
As described above, the plant cultivation management server 100 of this embodiment includes a target data acquisition unit 115 and an average size estimation unit 116. The target data acquisition unit 115 acquires target image data I(o), which is image data generated by photographing a cultivation section PL in which a plurality of target plants OP are cultivated. The average size estimation unit 116 estimates an average size AS, which is an average value of the sizes of each target plant OP cultivated in the cultivation section PL, using image data including color information of the target image data I(o) and an average size estimation model MO created by a predetermined machine learning.

In this way, the plant cultivation management server 100 of this embodiment can acquire the target image data I(o) generated by photographing the cultivation section PL, and estimate the average size AS of the target plant OP using image data including color information of the target image data I(o) and the average size estimation model MO. Therefore, according to the plant cultivation management server 100 of this embodiment, the average size AS of the target plant OP can be estimated simply by acquiring the target image data I(o), and the effort and time required to estimate the average size AS of the target plant OP can be reduced.

The plant cultivation management server 100 of this embodiment further includes a training data acquisition unit 112 that acquires training data TD in which a plurality of training image data I(t), which is image data generated by photographing a cultivation section PL in which a plurality of target plants OP are cultivated at a plurality of dates and times, and average size data indicating the average size AS of the target plants OP in the cultivation section PL represented by the training image data I(t), are associated with each other, and a model acquisition unit 114 that acquires an average size estimation model MO by machine learning using the training data TD. Therefore, according to the plant cultivation management server 100 of this embodiment, it is possible to acquire the average size estimation model MO without using any other device, and it is possible to estimate the average size AS of the target plants OP using the average size estimation model MO, thereby further reducing the effort and time required to estimate the average size AS of the target plants OP.

In addition, in the plant cultivation management server 100 of this embodiment, the training image data I(t) and the target image data I(o) include overhead image data Io, which is image data generated by photographing the cultivation section PL with a photographing device 300 (overhead photographing device 300A) in which the angle between the optical axis of the lens and the horizontal plane is 20 degrees or more. Therefore, according to the plant cultivation management server 100 of this embodiment, the average size AS can be estimated using an image in which the cultivation section PL is represented in a relatively wide image area, and the accuracy of estimating the average size AS of the target plant OP can be improved.

In addition, in the plant cultivation management server 100 of this embodiment, the training image data I(t) and the target image data I(o) include, in addition to the overhead image data Io, horizontal image data Ih, which is image data generated by photographing the cultivation section PL with a photographing device 300 (horizontal photographing device 300B) in which the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees. In addition, the model acquisition unit 114 acquires an average size estimation model MOo for the overhead view by machine learning using the training data TD for the overhead image data Io, and acquires an average size estimation model MOh for the horizontal view by machine learning using the training data TD for the horizontal image data Ih. The average size estimation unit 116 estimates the average size AS for the target image data I(o), which is the overhead image data Io, using the average size estimation model MOo for the overhead view, and estimates the average size AS for the target image data I(o), which is the horizontal image data Ih, using the average size estimation model MOh for the horizontal view. Therefore, according to the plant cultivation management server 100 of this embodiment, the average size AS can be estimated using a model optimized according to the image capture angle, and the accuracy of estimating the average size AS of the target plant OP can be further improved.

In addition, in the plant cultivation management server 100 of this embodiment, the training image data I(t) and the target image data I(o) are image data I generated by photographing the cultivation area PL during a preset daytime time. Therefore, according to the plant cultivation management server 100 of this embodiment, the average size AS can be estimated using an image in which the color of the target plant OP cultivated in the cultivation area PL is relatively faithfully represented, and the accuracy of estimating the average size AS of the target plant OP can be improved.

In addition, in the plant cultivation management server 100 of this embodiment, the training image data I(t) and the target image data I(o) are image data I generated by photographing an area including a plurality of cultivation sections PL and trimming the image data for each cultivation section PL. Therefore, according to the plant cultivation management server 100 of this embodiment, even if the growth conditions of the target plants OP differ for each cultivation section PL, the average size AS of the target plants OP cultivated in each cultivation section PL can be estimated with high accuracy. Note that the trimming at this time may be performed such that the image trimmed for one cultivation section PL does not include pixels representing the target plants OP cultivated in other cultivation sections PL, as in the above-mentioned trimming conditions 2 and 3. In this way, the influence of the pixels representing the target plants OP of the other cultivation sections PL can be eliminated, and the average size AS of the target plants OP cultivated in each cultivation section PL can be estimated with even higher accuracy.

The plant cultivation management server 100 of this embodiment further includes a harvest date and time prediction unit 117. The harvest date and time prediction unit 117 predicts the harvest date and time Te of the target plant OP based on the average size AS estimated for the target image data I(o). Therefore, according to the plant cultivation management server 100 of this embodiment, the harvest date and time Te of the target plant OP can be predicted simply by acquiring the target image data I(o), and the efficiency and optimization of plant cultivation can be effectively achieved.

In addition, in the plant cultivation management server 100 of this embodiment, the target data acquisition unit 115 acquires multiple target image data I(o) generated by photographing the target plant OP at multiple dates and times, and the average size estimation unit 116 estimates the average size AS for the multiple target image data I(o). In addition, the harvest date and time prediction unit 117 predicts the harvest date and time Te of the target plant OP for one target image data I(o) based on the average size AS estimated for the one target image data I(o) and the average size AS estimated for each target image data I(o) photographed before the photographing date and time of the one target image data I(o). Therefore, according to the plant cultivation management server 100 of this embodiment, the harvest date and time Te of the target plant can be predicted with high accuracy, and the efficiency and optimization of plant cultivation can be more effectively realized.

In addition, in the plant cultivation management server 100 of this embodiment, the target plant OP is a leafy vegetable, and the average size AS is the average plant height of the target plant OP. Therefore, according to the plant cultivation management server 100 of this embodiment, it is possible to reduce the effort and time required to estimate the average plant height of a leafy vegetable.

The plant cultivation management system 10 of this embodiment includes a plant cultivation management server 100 and a terminal device 200 capable of communicating with the plant cultivation management server 100 via a communication network NET. The plant cultivation management server 100 further includes an information transmission unit 118 that transmits estimated average size data SD indicating the estimated average size AS of the target plant OP to the terminal device 200 via the communication network NET. In this way, in the plant cultivation management system 10 of this embodiment, the information transmission unit 118 of the plant cultivation management server 100 transmits the estimated average size data SD indicating the estimated average size AS of the target plant OP to the terminal device 200, so that the farm worker holding the terminal device 200 can grasp the average size AS of the target plant OP even if he is at a location other than the plant factory PF, and can more effectively realize the efficiency and optimization of plant cultivation.

The plant cultivation management system 10 of this embodiment further includes a photographing device 300 that is installed in a field (e.g., a plant factory PF) where the target plant OP is cultivated and can communicate with the plant cultivation management system 10 via a communication network NET. The target data acquisition unit 115 of the plant cultivation management system 10 acquires the target image data I(o), which is image data I generated by the photographing device 300 photographing the cultivation section PL, via the communication network NET. Therefore, in the plant cultivation management system 10 of this embodiment, the photographer (e.g., a farm worker) can acquire the target image data I(o) without going to the field, and the average size AS of the target plant OP can be estimated using the acquired target image data I(o), thereby making it possible to achieve extremely effective efficiency and optimization of plant cultivation.

A-5. Example:
An example performed to verify the accuracy of the estimation of the average size AS by the plant cultivation management server 100 of the present embodiment described above will be described below. FIG. 12 is an explanatory diagram showing the prediction accuracy of the average size AS for each average size estimation model MO. FIG. 12 shows the value of the coefficient of determination (R ² ) as an index of the prediction accuracy of the average size AS for each average size estimation model MO used to estimate the average size AS. Example E1 is an example in which the average size AS is estimated using the average size estimation model MOo for the bird's-eye view only for the bird's-eye view image data Io, Example E2 is an example in which the average size AS is estimated using the average size estimation model MOh for the horizontal image data Ih only, and Example E3 is an example in which the average size AS is estimated using the average size estimation model MOo for the bird's-eye view for the bird's-eye view image data Io, and the average size AS is estimated using the average size estimation model MOh for the horizontal image data Ih. Moreover, Example E4 is an example in which an average size AS for target image data I(o) containing a mixture of overhead image data Io and horizontal image data Ih is estimated using a common average size estimation model MO created using training data TD containing a mixture of overhead image data Io and horizontal image data Ih without distinguishing between overhead and horizontal images. Note that in each example, only image data I taken between 7:00 and 17:00 was targeted, and images cropped according to the above-mentioned cropping condition 1 were used. In addition, machine learning for creating the average size estimation model MO used VGG16, and the prediction results of five basic learners were simply averaged according to the parallel ensemble method.

Comparing Example E3 and Example E4, Example E3 has a higher estimation accuracy of the average size AS. Therefore, it is preferable to estimate the average size AS for the overhead image data Io using the overhead average size estimation model MOo and to estimate the average size AS for the horizontal image data Ih using the horizontal average size estimation model MOh, rather than estimating the average size AS for the target image data I(o) in which overhead image data Io and horizontal image data Ih are mixed using a common average size estimation model MO created using training data TD in which overhead image data Io and horizontal image data Ih are mixed. The reason for this is that if the overhead average size estimation model MOo is created using only the overhead image data Io, the model becomes a model specialized for the overhead image data Io, and the accuracy of estimating the average size AS from the overhead image data Io improves (the same applies to the horizontal image data Ih).

Furthermore, comparing Example E1 and Example E2, Example E1 has a higher estimation accuracy for the average size AS. Therefore, it can be said that it is preferable to create the average size estimation model MO and estimate the average size AS using the overhead image data Io rather than using the horizontal image data Ih. The reason for this is thought to be that the overhead image data Io is an image in which the cultivation section PL is shown in a relatively wide image area, and is an image that clearly shows the characteristics related to the size of the target plant OP being cultivated in the cultivation section PL.

FIG. 12 is an explanatory diagram showing the prediction accuracy of the average size AS for each trimming condition. FIG. 13 shows the value of the coefficient of determination (R ² ) as an index of the prediction accuracy of the average size AS for each trimming condition for the image data I. Example E11 is an example in which the average size estimation model MO is created and the average size AS is estimated using image data I that has been trimmed according to the above-mentioned trimming condition 1, Example E12 is an example in which the average size estimation model MO is created and the average size AS is estimated using image data I that has been trimmed according to the above-mentioned trimming condition 2, and Example E13 is an example in which the average size estimation model MO is created and the average size AS is estimated using image data I that has been trimmed according to the above-mentioned trimming condition 3. In each example, only image data I taken during the time period from 7:00 to 17:00 was targeted. In addition, machine learning for creating the average size estimation model MO used VGG16, and the prediction results of five basic learners were simply averaged according to the parallel ensemble method. In addition, for the overhead image data Io, the average size AS is estimated using the overhead average size estimation model MOo, and for the horizontal image data Ih, the average size AS is estimated using the horizontal average size estimation model MOh.

Comparing Examples E11, E12, and E13, the average size AS is more accurately estimated in Examples E12 and E13 than in Example E11. It can be said that it is more preferable to create an average size estimation model MO and estimate an average size AS using image data I cropped according to cropping condition 2 or cropping condition 3 than to create an average size estimation model MO and estimate an average size AS using image data I cropped according to cropping condition 1. Cropping condition 2 and cropping condition 3 are cropping conditions that are executed so that an image cropped for one cultivation section PL does not include pixels representing target plants OP cultivated in other cultivation sections PL. Therefore, if cropping conditions that are executed so that an image cropped for one cultivation section PL does not include pixels representing target plants OP cultivated in other cultivation sections PL are adopted, it can be said that the average size AS of the target plants OP cultivated in each cultivation section PL can be estimated with even higher accuracy.

B. Variations:
The technology disclosed in this specification is not limited to the above-described embodiments, and can be modified in various forms without departing from the spirit of the invention. For example, the following modifications are also possible.

The configuration of the plant cultivation management system 10 in the above embodiment is merely an example and can be modified in various ways. For example, in the above embodiment, the plant cultivation management system 10 is equipped with the photographing device 300, but the plant cultivation management system 10 does not have to be equipped with the photographing device 300. Even if the plant cultivation management system 10 does not have the photographing device 300, for example, a farm worker can use a photographing device (e.g., a camera mounted on the terminal device 200) to photograph a plant cultivated in the plant factory PF to generate image data, and transmit the image data to the plant cultivation management server 100, so that the plant cultivation management server 100 can obtain image data of the plant.

In addition, in the above embodiment, the photographing device 300 transmits image data generated by photographing to the plant cultivation management server 100 each time photographing is performed, but the image data may be transmitted to the plant cultivation management server 100 at other times (for example, when a request is received from the plant cultivation management server 100).

In the above embodiment, the plant cultivation management server 100 executes an average size estimation model creation process to create the average size estimation model MO, but the average size estimation model creation process may be executed by a device other than the plant cultivation management server 100, and the model acquisition unit 114 of the plant cultivation management server 100 may acquire the average size estimation model MO from the other device. Alternatively, the plant cultivation management server 100 may transmit the target image data I(o) to the other device, and the plant cultivation management server 100 may acquire the average size AS estimated by the other device using the average size estimation model MO.

The contents of the plant cultivation support process in the above embodiment are merely examples and can be modified in various ways. For example, in the plant cultivation support process in the above embodiment, estimation of the average size AS (S220) and prediction of the harvest date and time Te (S240) are performed, but it is also possible to only estimate the average size AS and not predict the harvest date and time Te.

In the above embodiment, the prediction of the harvest date and time Te from the average size AS is performed by simple regression, but the prediction of the harvest date and time Te may also be performed by other methods (e.g., multiple regression).

In addition, in each of the above embodiments, some of the configurations realized by hardware may be replaced by software, and conversely, some of the configurations realized by software may be replaced by hardware.

10: Plant cultivation management system 100: Plant cultivation management server 110: Control unit 111: Plant cultivation management unit 112: Training data acquisition unit 114: Model acquisition unit 115: Target data acquisition unit 116: Average size estimation unit 117: Harvest date and time prediction unit 118: Information transmission unit 120: Memory unit 130: Display unit 140: Operation input unit 150: Interface unit 190: Bus 200: Terminal device 210: Display unit 300: Photographing device 300A: Overhead photographing device 300B: Horizontal photographing device CP: Plant cultivation management program GD: Estimated average size data MO: Average size estimation model MOh: Horizontal average size estimation model MOo: Overhead average size estimation model NET: Communication network OP: Target plant PD: Estimated harvest date and time data PF: Plant factory PL: Cultivation area SD: Estimated average size data TD: Training data TF: Trimming frame

Claims (11)

An information processing device,
a target data acquisition unit that acquires target image data, which is image data generated by photographing a cultivation section in which a plurality of target plants, which are a specific type of plant, are cultivated;
an average size estimation unit that estimates an average size, which is an average value of the sizes of the target plants cultivated in the cultivation area, by using image data including color information of the target image data and a model created by a predetermined machine learning method;
a training data acquisition unit that acquires training data in which a plurality of training image data, which is image data generated by photographing a cultivation section in which a plurality of target plants are cultivated at a plurality of dates and times, and average size data indicating an average size of the target plants in the cultivation section represented by the training image data are associated with each other;
a model acquisition unit that acquires the model by the machine learning using the training data;
Equipped with
The training image data and the target image data include overhead image data, which is image data generated by photographing the cultivation area with a photographing device having an angle between the optical axis of a lens and a horizontal plane of 20 degrees or more,
The training image data and the target image data include, in addition to the overhead image data, horizontal image data, which is image data generated by photographing the cultivation area with a photographing device in which the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees,
the model acquisition unit acquires a first model by the machine learning using the training data for the overhead image data, and acquires a second model by the machine learning using the training data for the horizontal image data;
the average size estimation unit estimates the average size of the target image data, which is the overhead image data, using the first model, and estimates the average size of the target image data, which is the horizontal image data, using the second model .
Information processing device. 2. The information processing device according to claim 1 ,
The training image data and the target image data are image data generated by photographing the cultivation area during a preset daytime time.
Information processing device. 3. The information processing device according to claim 1 ,
The training image data and the target image data are image data generated by photographing an area including a plurality of the cultivation sections and trimming each of the cultivation sections.
Information processing device. 4. The information processing device according to claim 3 ,
The cropping is performed such that a cropped image of one of the cultivation sections does not include pixels representing the target plants being cultivated in another of the cultivation sections.
Information processing device. The information processing device according to any one of claims 1 to 4 , further comprising:
a harvest date prediction unit that predicts a harvest date and time of the target plant based on an average size of the target plant estimated for the target image data;
Information processing device. 6. The information processing device according to claim 5 ,
The target data acquisition unit acquires a plurality of target image data generated by photographing the cultivation area at a plurality of dates and times,
The average size estimation unit estimates an average size of the target plant for a plurality of the target image data;
the harvest date and time prediction unit predicts a harvest date and time of the target plant for one of the target image data based on an average size of the target plant estimated for the one of the target image data and an average size of the target plant estimated for each of the target image data photographed before the photographing date and time of the one of the target image data;
Information processing device. The information processing device according to any one of claims 1 to 6 ,
The target plant is a leafy vegetable,
The average size is the average plant height of the target plant;
Information processing device. An information processing device according to any one of claims 1 to 7 ;
a terminal device capable of communicating with the information processing device via a communication network;
An information processing system comprising:
The information processing device further includes an information transmitting unit that transmits information indicating the estimated average size of the target plant to the terminal device via the communication network.
Information processing system. The information processing system according to claim 8 , further comprising:
a photographing device that is installed in a field where the target plant is cultivated and is capable of communicating with the information processing device via a communication network;
The target data acquisition unit acquires the target image data generated by the photographing device photographing the cultivation area via the communication network.
Information processing system. 1. An information processing method, comprising:
acquiring target image data, the target image data being image data generated by photographing a cultivation plot in which a plurality of target plants of a particular type are cultivated;
A step of estimating an average size, which is an average value of the sizes of each of the target plants cultivated in the cultivation area, using image data including color information of the target image data and a model created by a predetermined machine learning method;
acquiring training data in which a plurality of training image data, which is image data generated by photographing a cultivation section in which a plurality of target plants are cultivated, at a plurality of dates and times, is associated with average size data indicating an average size of the target plants in the cultivation section represented by the training image data;
obtaining the model by the machine learning using the training data;
Equipped with
The training image data and the target image data include overhead image data, which is image data generated by photographing the cultivation area with a photographing device having an angle between the optical axis of a lens and a horizontal plane of 20 degrees or more,
The training image data and the target image data include, in addition to the overhead image data, horizontal image data, which is image data generated by photographing the cultivation area with a photographing device in which the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees,
the step of acquiring the model includes a step of acquiring a first model by the machine learning using the training data for the overhead image data, and acquiring a second model by the machine learning using the training data for the horizontal image data;
the step of estimating the average size is a step of estimating the average size of the target image data, which is the overhead image data, using the first model, and estimating the average size of the target image data, which is the horizontal image data, using the second model;
Information processing methods. A computer program comprising:
On the computer,
acquiring target image data, the target image data being image data generated by photographing a cultivation plot in which a plurality of target plants of a particular type are cultivated;
A process of estimating an average size, which is an average value of the sizes of each of the target plants cultivated in the cultivation area, using image data including color information of the target image data and a model created by a predetermined machine learning method;
A process of acquiring training data in which a plurality of training image data, which is image data generated by photographing a cultivation section in which a plurality of target plants are cultivated, at a plurality of dates and times, and average size data indicating an average size of the target plants in the cultivation section represented by the training image data are associated with each other;
obtaining the model by the machine learning using the training data;
Run the command ,
The training image data and the target image data include overhead image data, which is image data generated by photographing the cultivation area with a photographing device having an angle between the optical axis of a lens and a horizontal plane of 20 degrees or more,
The training image data and the target image data include, in addition to the overhead image data, horizontal image data, which is image data generated by photographing the cultivation area with a photographing device in which the angle between the optical axis of the lens and the horizontal plane is less than 20 degrees,
the process of acquiring the model is a process of acquiring a first model by the machine learning using the training data for the overhead image data, and acquiring a second model by the machine learning using the training data for the horizontal image data,
The process of estimating the average size is a process of estimating the average size of the target image data, which is the overhead image data, using the first model, and estimating the average size of the target image data, which is the horizontal image data, using the second model .
Computer program.

Images