JP2023152095A - Crop growth state estimation device, and crop growth state estimation method - Google Patents

Abstract

An object of the present invention is to accurately determine growth conditions based on a variety of characteristic amounts.
A target crop feature acquisition unit that acquires a target crop feature that is a feature related to the target crop that is currently being cultivated; a target crop feature storage unit that stores the target crop feature; A probability calculation unit that calculates the probability of the growth state using a Bayesian network model based on the target agricultural product feature quantity, and a growth state display screen that displays the estimated growth state according to the calculated probability. What is claimed is: 1. A crop growth state estimating device comprising: a screen generating section for generating a screen; and a display section.
[Selection diagram] Figure 1

Description

The present invention relates to a crop growth state estimation device and a crop growth state estimation method.

In recent years, as part of smart agriculture, there has been an increasing demand for systems that monitor the growth status of agricultural crops. Detecting and predicting the occurrence of pests and physiological disorders in agricultural crops is one of the functions required of such systems. Some pests and physiological disorders that occur in agricultural crops have similar appearances, and it is difficult to determine whether they are pests or physiological disorders using image recognition alone. Regarding pests and physiological disorders, techniques have been proposed for predicting the occurrence of at least one of pests and physiological disorders (see Patent Document 1) and techniques for estimating locations where pests are likely to occur (see Patent Document 2). Neither of these methods distinguishes between pests and physiological disorders.

JP2021-93957A Japanese Patent Application Publication No. 2015-119646

However, each method requires different treatments, such as spraying pesticides for pests and adding fertilizers and nutrients for menstrual disorders. It is very important to distinguish between

The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to accurately determine the growth state of agricultural products based on a variety of characteristic amounts.

The present invention for solving the above problems is as follows:
a target crop feature amount acquisition unit that acquires a target crop feature amount that is a feature amount regarding the target crop that is currently cultivated;
a target crop feature amount storage unit that stores the target crop feature amount;
a probability calculation unit that calculates the probability of the growth state of the target crop using a Bayesian network model based on the target crop feature;
a screen generation unit that generates a growth state display screen that displays the estimated growth state according to the calculated probability;
A display section;
This is an agricultural crop growth state estimation device characterized by comprising:

According to this, the estimated growth state is displayed on the display section according to the probability calculated by a Bayesian network model based on the target crop feature amount that is the feature amount related to the target crop that the user is currently cultivating. . For this reason, the crop growth state estimating device uses a Bayesian network model that probabilistically estimates the growth state of crops based on feature amounts acquired from growth-related information related to the growth of the crops. It is possible to accurately determine the growth status of crops based on the amount, including whether the abnormality is due to pests or physiological disorders, and shows the user the growth status of the crops currently being cultivated. Since the information is displayed according to the probability, the growth status of the crops currently being cultivated can be clearly recognized visually.

Furthermore, in the present invention,
The screen generation unit includes:
A growth state ranking display screen may be generated that displays the estimated growth state in association with the calculated probability ranking.

According to this, by using a Bayesian network model, the crop growth state estimating device can accurately determine the growth state of crops based on a variety of feature quantities, and the user can easily determine the growth state of the crops currently being cultivated. Since the display shows which growth state is most likely to be present, the growth state of the crops currently being cultivated can be clearly recognized visually.

Furthermore, in the present invention,
A teacher data storage unit that stores teacher data in which a feature amount acquired from growth-related information related to the growth of the agricultural product is associated with growth state information indicating the growth state of the agricultural product targeted by the growth-related information. and,
a Bayesian network model construction unit that constructs a Bayesian network model using the teacher data stored in the teacher data storage unit;
It is equipped with a Bayesian network model generation unit having
The probability calculation unit may calculate the probability of the growth state using the Bayesian network model generated by the Bayesian network model generation unit.

In this way, the growth state estimating device can model the causal relationship between the feature amount and the growth state using a network structure.

Furthermore, in the present invention,
The Bayesian network model generation unit includes:
The device may include a Bayesian network model adjustment unit that changes at least one of nodes, edges, and probability values that constitute the Bayesian network model constructed by the Bayesian network model construction unit.

In this way, in addition to the information prepared as training data, if the known know-how regarding cultivation of agricultural products is possessed, the nodes constituting the Bayesian network model constructed by the Bayesian network model construction section, By changing at least either the edges or the probability values, a Bayesian network model that reflects the know-how can be generated.

Furthermore, in the present invention,
The screen generation unit includes:
A graph display screen that displays the graph structure of the Bayesian network model may be generated.

In this way, the user can understand the relationship structure between the feature amounts and the growth state that constitute the nodes of the Bayesian network model, that is, the causal relationship between them.

Moreover, the present invention
a step of acquiring a target crop feature amount that is a feature amount related to the target crop that is currently being cultivated;
a step of storing the target agricultural product feature quantity;
Calculating the probability of the growth state of the target crop using a Bayesian network model based on the target crop feature amount;
generating a growth state display screen that displays the estimated growth state according to the calculated probability;
Displaying the growth status display screen;
This is a method for estimating the growth state of agricultural crops.

According to this, the estimated growth state is displayed according to the probability calculated by the Bayesian network model based on the target crop feature amount which is the feature amount regarding the target crop currently being cultivated by the user. Therefore, according to crop growth state estimation, by using a Bayesian network model that probabilistically estimates the growth state of agricultural products based on the feature values obtained from growth-related information related to the growth of agricultural products, it is possible to It is possible to accurately determine the growth state of crops based on feature values, including whether the abnormality is caused by pests or physiological disorders, and allows users to know the growth state of the crops they are currently cultivating. Since the information is displayed according to the probability shown, the growth status of the crops currently being cultivated can be clearly recognized visually.

According to the present invention, it is possible to accurately determine the growth state of agricultural products based on a variety of characteristic amounts.

1 is a schematic configuration diagram of a crop growth state observation system according to an embodiment of the present invention. 1 is a schematic configuration diagram of a cultivation facility to which a crop growth state observation system according to an embodiment of the present invention is applied. It is a figure showing the data structure of agricultural product information TSDB concerning an example of the present invention. FIG. 2 is a diagram showing a data structure of a pest and physiological disorder information DB according to an embodiment of the present invention. It is a figure showing the data structure of sensor data TSDB concerning an example of the present invention. FIG. 3 is a diagram showing a data structure of public data TSDB according to an embodiment of the present invention. FIG. 3 is a diagram showing a data structure of image data TSDB according to an embodiment of the present invention. It is a figure showing the data structure of judgment result TSDB concerning an example of the present invention. FIG. 1 is a functional block diagram of a crop growth state observation device according to an embodiment of the present invention. It is a flowchart explaining the procedure of the agricultural crop growth state observation method based on the Example of this invention. It is a graph explaining discretization processing concerning an example of the present invention. FIG. 2 is a diagram illustrating construction of a Bayesian network according to an embodiment of the present invention. FIG. 2 is a diagram illustrating estimation by a Bayesian network according to an embodiment of the present invention. FIG. 2 is a diagram illustrating estimation by a Bayesian network according to an embodiment of the present invention. FIG. 3 is a diagram illustrating a display example of a Bayesian network according to an embodiment of the present invention. FIG. 2 is a diagram illustrating estimation by a Bayesian network according to an embodiment of the present invention. It is a figure which shows the example of an occurrence probability ranking screen obtained by the Bayesian network based on the Example of this invention.

[Application example]
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A crop growth state observation system 1 according to an application example of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing a schematic configuration of a crop growth condition observation system 1, and FIG. 2 is a diagram showing a schematic configuration of a crop cultivation facility 200 to which the crop growth condition observation system 1 is applied.
The agricultural product growth state observation system 1 observes the growth state of the agricultural products 100 cultivated in the cultivation facility 200 based on the camera 40, the sensor 50, public data, and the like. Regarding the growth condition of crops, it may be difficult to distinguish from their appearance, such as gray mold and calcium deficiency. Treatment methods for improving symptoms vary depending on the cause, such as whether the symptoms are caused by pests or menstrual disorders, so in order to respond appropriately, it is necessary to accurately distinguish between these symptoms. be.

The crop growth state observation device 10 included in the crop growth state observation system 1 is composed of functional blocks shown in FIG. This agricultural product growth state observation device 10 uses data acquired through a camera 40, a sensor 50, and a public data acquisition unit 60, or features extracted from these data, and judgment results regarding the growth state (for example, normal, gray mold, etc.). A Bayesian network 400 that estimates the probability of occurrence of symptoms (growth conditions) of gray mold, etc. from input data such as values of the sensor 50, as shown in FIG. generate.

In the crop growth state observation system 1, the data acquired by the camera 40, the sensor 50, the public data acquisition unit, and the feature amounts extracted from these data are input to the Bayesian network 400, so that the growth state of the crops is determined to be normal. , gray mold, and calcium deficiency are calculated, and as shown in FIG. 17, a plurality of symptoms can be displayed in a ranking format according to the high probability of occurrence.

In this way, by using the Bayesian network 400 and comprehensively considering information from various data, it is possible to accurately determine the causes of pests and menstrual disorders with similar symptoms. It is possible to detect and predict occurrences and take appropriate measures.

[Example 1]
EMBODIMENT OF THE INVENTION Hereinafter, the structure of the agricultural crop growth state observation system 1 based on Example 1 of this invention is demonstrated with reference to drawings. However, the configuration of the apparatus described in this example should be modified as appropriate depending on various conditions. That is, the scope of the present invention is not intended to be limited to the following examples.

FIG. 1 shows a schematic configuration of a crop growth state observation system 1 according to this embodiment.
The crop growth state observation system 1 includes a crop growth state observation device 10, a data management server 20, a growth state display device 30, a camera 40, a sensor 50, and a public data acquisition unit 60.

The camera 40 is an imaging device that photographs the state of the agricultural products 100. The camera 40 can be configured by, for example, a fixed-point observation camera, a camera built into a smartphone, a multispectral camera, a camera mounted on a drone, or an orbiting robot, but is not limited thereto.

The sensor 50 includes a first sensor 50a, a second sensor 50b,..., an n-th sensor 50c. These sensors 50 are a group of sensors that measure environmental factors that affect the growth of agricultural products, and include, but are not limited to, soil sensors, temperature/humidity sensors, sunlight sensors, CO2 sensors, soil PH sensors, etc. do not have. When referring to a group of sensors collectively, they are simply referred to as sensors 50.

The public data acquisition unit 60 includes a first public data acquisition unit 60a, a second public data acquisition unit 60b, . . . , an m-th public data acquisition unit 60c. This is an acquisition unit that acquires publicly available data that can be used in the agricultural field. The public data acquisition unit 60 uses a communication device that acquires public data from an external server connected via a network, a reading device that reads public data provided on a recording medium, etc., depending on the provision form of each public data. configured, but not limited to this. Publicly available data includes, but is not limited to, mesh agricultural meteorological data, national digital soil maps, pest outbreak forecast information, etc., for example. When referring to a plurality of public data acquisition units collectively, they are simply referred to as the public data acquisition unit 60.

The agricultural product growth state observation device 10 uses captured image data received from the camera 40, sensor data measured by the sensor 50, public data acquired by the public data acquisition unit 60, etc. to determine the growth state of the agricultural products to be observed. It is a device for estimating. Examples of the growing state of agricultural crops include, but are not limited to, a state where the crop is affected by pests, a state where a physiological disorder has occurred, and the like. The crop growth state observation device 10 includes, for example, a CPU (processor), a main storage device (memory), an auxiliary storage device (hard disk, etc.), an input device (keyboard, mouse, controller, touch panel, etc.), and an output device (printer, speaker, etc.). ), it can be configured as a general-purpose computer system equipped with a communication device. Here, information related to the growth of agricultural products including photographed image data received from the camera 40, sensor data measured by the sensor 50, public data acquired by the public data acquisition unit 60, etc. is the growth related information of the present invention. corresponds to The growth related information of the present invention is not limited to these.

The growth state display device 30 is a device that displays information such as processing results of the crop growth state observation device 10 and images taken by the camera 40, and can be configured by an output device such as a display.

The data management server 20 is a server that manages various databases (DB), and includes, for example, a CPU (processor), main storage (memory), auxiliary storage (hard disk, etc.), input devices (keyboard, mouse, controller, touch panel, etc.). ), an output device (display, printer, speaker, etc.), and a communication device.
The DBs managed by the data management server 20 include a crop information DB 20a, a sensor data TSDB 20b, a public data TSDB 20c, a pest and physiological disorder information DB 20d, an image data TSDB 20e, and a determination result TSDB 20f.
The agricultural product information DB 20a is a database that stores information regarding agricultural products whose growth conditions are to be observed.
The sensor data TSDB 20b is a database that stores sensor data, which is data measured by the sensor 50, in chronological order.
The public data TSDB 20c is a database that stores data acquired by the public data acquisition unit 60 in chronological order.
The pest and disease/physiological disorder information DB 20d is a database that stores information regarding pests and physiological disorders of agricultural products.
The image data TSDB 20e is a database that stores image data taken by the camera 40 in chronological order.
The determination result TSDB 20f is a database that stores in chronological order the determination results of whether symptoms appearing on agricultural products are caused by pests or physiological disorders.

FIG. 2 is a diagram schematically showing a schematic configuration of a cultivation facility 200 such as a greenhouse in which crops to be observed by the above-mentioned crop growth state observation system 1 are cultivated.

In the cultivation facility 200, agricultural products 100 are cultivated in three rows: A row, B row, and C row. Row A is provided with six cultivation sections A1 to A6, row B is provided with six cultivation sections B1 to B6, and row C is provided with six cultivation sections C1 to C6.

A crop ID for identification is assigned to each crop cultivated in each cultivation section, for example, a crop ID such as 000001 is assigned. Cultivated in the cultivation facility 200 in this way

A camera 40 for photographing the growth state of agricultural crops and a sensor 50 for measuring various environmental factors are arranged. FIG. 2 shows an example in which the camera 40, the first sensor 50a, and the second sensor 50b are arranged. The sensors 50 are also given sensor IDs for identifying each sensor 50. For example, the first sensor 50a is given a sensor ID of 1A001, and the second sensor 50b is given a sensor ID of 2A001.

In FIG. 2, crops cultivated in the A1 cultivation section and A2 cultivation section are illustrated, but crops similarly assigned crop IDs are also shown in other cultivation sections A3 to A6, B1 to B6, and C1 to C6. An appropriate number of cameras 40 and sensors 50 are arranged. When a plurality of cameras 40 are installed or used in the cultivation facility 200, a camera ID for identifying each camera 40 is given.

FIG. 3 is a diagram schematically showing an example of the data structure of the agricultural product information DB 20a.
The agricultural product information DB 20a includes, for example, 000001 as the agricultural product ID, A1 as the cultivation location, tomato as the name, 1A001 as the first sensor ID, 2A001 as the second sensor ID, PB1_001 as the first draft or data ID, and the second public data ID. Pb2_001 is stored as the image data ID, IM_000001 is stored as the image data ID, and 138.0036.00, 135.00 35.00 is stored as the GPS position information. Here, the GPS position information is information for associating which crop ID an image photographed with a smartphone or the like corresponds to. Further, the agricultural product ID may be linked to information obtained from an individual identification information tag such as a QR code (registered trademark) or RFID attached to the agricultural product.

The crop information DB 20a may store information regarding drugs, instruments, devices, etc. that are provided for use in cultivating the crop.

FIG. 4 is a diagram schematically showing an example of the data structure of the pest and physiological disorder information DB 20d.
The pest and physiological disorder information DB 20d stores, for example, information such as symptom names, pests or physiological disorders, symptom image information, explanations of symptoms, explanations of causes, frequent occurrence periods, countermeasures, and usable drugs. Specifically, the name of a symptom such as botrytis is stored as a symptom name, the distinction between a pest or a physiological disorder (in this case, a pest) is stored as a symptom image information, and "Hairokabibyyou" is stored as a symptom image information. ．． jpg and the file name of the symptom image information are stored. In addition, the symptoms are explained as follows: ``A gray mold develops in a water-stained lesion on the calyx of the fruit.The fruit is attacked from around the calyx, and gray mold grows, causing it to wither, die, and rot. occurs. Yellowish-white spots with a diameter of 1 to 2 mm often appear on the fruit.'' is memorized. The explanation for the cause is as follows: ``The fungus that causes gray mold prefers a humid environment, so it is more likely to develop on lower parts of plants that are planted without sufficient distance between them, or on plants that have a lot of leaves and stems.'' The explanatory text will be memorized. Also, the period from April to June is remembered as the period when outbreaks occur frequently, and the countermeasures are to ``improve ventilation and lighting, and reduce humidity by using ventilation and humidifiers.After an outbreak occurs, immediately wrap it in a bag and take it outdoors. ” is stored, and the names of drugs such as drug A, drug B, drug C, etc. are stored as usable drugs.

FIG. 5 is a diagram schematically showing a data configuration example of the sensor data TSDB 20b. In the sensor data TSDB 20b, sensor data for each sensor specified by the sensor ID is stored in association with the date and time of data acquisition. For example, the sensor data is stored as 20.9 for a sensor with a sensor ID of 1A001 acquired at 00:00:00 on January 6, 2022, and 45.3 for a sensor with a sensor ID of 2A001. be done. Here, the sensor data may not be raw measured values but feature amounts obtained by processing them.

FIG. 6 is a diagram schematically showing an example of the data structure of the public data TSDB 20c. In the public data TSDB 20c, data for each public data identified by the public data ID is stored in association with data acquisition date and time. For example, public data is stored as 1321.3 for data with public data ID Pb1_001 acquired at 00:00:00 on January 6, 2022, and 1 (many) for data with public data ID Pb2_001. Ru. Here, the data whose public data ID is Pb1_001 is precipitation, and the data whose public data ID is Pb2_001 is pest incidence information, but the contents of the public data are not limited to these. Here, the data acquisition date and time includes the date and time that is the subject of the public data, depending on the content of the public data. For example, for precipitation, the date and time when the precipitation was measured is the data acquisition date and time, and for pest incidence information, the data acquisition date and time is the date and time when pest outbreaks are observed or predicted. There may be.

FIG. 7 is a diagram schematically showing an example of the data structure of the image data TSDB 20e. In the image data TSDB 20e, the file name of the image data specified by the image data ID and the feature amount 1 extracted from the image data are stored in association with the acquisition date and time for the agricultural product specified by the agricultural product ID. There is. For example, for a crop whose crop ID is 000001 acquired at 00:00:00 on January 6, 2022, the file name of the image data whose image data ID is IM_000001 is 2022_01_06_00_00_00_IM_000001. jpg, and the first feature amount of the image data is 561.3.

FIG. 8 is a diagram schematically showing an example of the data structure of the determination result TSDB 20f. In the judgment result TSDB 20f, the cultivation location, the sensor data of the first sensor 50a to the nth sensor 50c, the data of the first public data to the mth public data, the first image recognition feature amount to the kth image recognition feature amount, and the determination result are It is stored in association with the judgment date and time. As the cultivation location, a cultivation section name such as A1 is stored. Since the sensor data and the public data have been described above, their explanation will be omitted. The image recognition feature amount is a feature amount extracted by processing the image data described above. Further, as the determination results, normal, botrytis, which is a disease and pest, and calcium deficiency and calcium excess, which are physiological disorders, are stored.

Here, gray mold, calcium deficiency, and calcium excess are determined as examples of pests and physiological disorders that have similar symptoms, but pests and physiological disorders that have similar symptoms are limited to these. do not have. For example, ring spot disease caused by pests and potassium deficiency due to excessive phosphate intake, which is a physiological disorder, and tomato yellowing disease caused by pests and diseases, and magnesium deficiency, which is a physiological disorder, have similar symptoms, so they should be differentiated. It may be determined separately.

FIG. 9 shows a functional block diagram of the growth state observation device 10. The growth state observation device 10 mainly includes a Bayesian network generation section 11, a Bayesian network storage section 12, and a growth state estimation section 13. The specific functions of each part will be explained with reference to the flowchart shown in FIG. Here, the Bayesian network generation section 11 corresponds to the Bayesian network model generation section of the present invention. The crop growth state estimation device of the present invention includes a crop growth state observation device 10, a data management server 20, and a growth state display device 30. Hereinafter, the Bayesian network model will also be simply referred to as a Bayesian network.

The Bayesian network generation unit 11 is a functional unit that generates a Bayesian network used for estimating and predicting various information regarding the growth status of crops to be cultivated. Function is realized. The Bayesian network generation section 11 includes a training data acquisition section 11a, a feature value calculation section 11b, a determination information generation section 11c, a discretization processing section 11d, a Bayesian network construction section 11e, and a Bayesian network adjustment section 11f. Here, the Bayesian network construction section 11e corresponds to the Bayesian network model construction section of the present invention, and the Bayesian network adjustment section 11f corresponds to the Bayesian network model adjustment section of the present invention.

The Bayesian network storage unit 12 is a functional unit that stores the Bayesian network generated by the Bayesian network generation unit 11, and is configured by, for example, an auxiliary storage device.

The growth state estimating unit 13 is a functional unit that estimates and predicts various information regarding the growth state of agricultural products using the Bayesian network stored in the Bayesian network storage unit 12, and a predetermined program is executed by the CPU. Each function is realized by The growth state estimation section 13 includes a data acquisition section 13a, a feature value calculation section 13b, an input data acquisition section 13c, a probability calculation section 13d, and an output information generation section 13e.

FIG. 10 shows a flowchart illustrating the procedure of the crop growth state observation method executed in the crop growth state observation system 1.
This crop growth state observation method mainly includes the steps of generating a Bayesian network for estimating and predicting the growth state of crops, and using the generated Bayesian network to provide information useful for crop growth. include.

<Generation of Bayesian network>
First, various sensors 50 are installed in the cultivation sections A1 to A6, B1 to B6, and C1 to C6 of the cultivation facility 200 illustrated in FIG. 2 (step S1).

Next, information regarding cultivated crops, etc. is registered in the crop information DB 20a (step S2).

Next, data or feature amounts are collected from the sensor data of the installed sensor 50, the public data acquired from the public data acquisition unit 60, and the image data photographed by the camera 40, and the sensor data TSDB 20b, public data TSDB 20c, Each image data is stored in the TSDB 20e (step S3). Here, sensor data, public data, and image data are acquired by the training data acquisition unit 11a of the growth state observation device 10, and if necessary, the raw data is subjected to feature calculation processing in the feature calculation unit 11b. You may calculate the feature amount by executing In this way, the acquired sensor data, etc. or feature quantities are stored in the sensor data TSDB 20b, the public data TSDB, and the image data TSDB 20e. In the following, data or feature amounts are also collectively referred to as feature amounts.

Next, each data is labeled with a determination result (growth state information) such as normal, botrytis, calcium deficiency, etc. (step S4). Each piece of data labeled with a determination result is stored in the determination result TSDB 20f in a data configuration as shown in FIG. To label the determination results, the determination information generation unit 11c of the crop growth state observation device 10 reads the data stored in the sensor data TSDB 20b, public data TSDB, and image data TSDB 20e to the crop growth state observation device 10, and displays the growth state. The data is displayed on the device 30, input of the determination result is accepted from the user through the input device, and the acquired determination result is associated with these data and stored in the determination result TSDB 20f.

Next, each piece of data obtained as continuous values is discretized (step S5). For example, the data or feature amount extracted from the first sensor 50a is a continuous value and shows a distribution as shown in FIG. 11, and the range of the data or feature amount extracted from the first sensor 50a is from 0.0 to 60. Suppose it is 0. At this time, the range of the sensor data can be discretized at intervals of 10.0 using discretization (EWD (Equal Width Discretization)) that divides the range into equally spaced intervals. The discretization of data or feature amounts may be performed by the discretization processing unit 11d of the crop growth state observation device 10 acquiring the range of the sensor etc. in advance and setting it based on this, or by using the graph shown in FIG. Necessary information may be provided by displaying it on the growth state display device 30, and settings by the user such as sections for discretization may be accepted through the input device. The discretization interval is not limited to 10.0, but any appropriate interval can be adopted. In addition, the method of discretizing each data is not limited to this, but includes discretization using frequency intervals (EFD (Equal Frequency Discretization)), chimerge, CAIM, discretization using the minimum description length principle (MDLP (Minimum Description Length Principle)). ) etc. may be adopted. The discretized data is stored in the determination result TSDB 20f by the discretization processing unit 11d. Data in which the discretized feature amounts and determination results are associated constitutes training data for learning the Bayesian network. Here, the training data corresponds to the teacher data of the present invention, and the determination result TSDB corresponds to the teacher data storage unit of the present invention.

Next, the Bayesian network construction unit 11e of the crop growth state observation device 10 constructs a Bayesian network model from the discretized data (step S6).

FIG. 12 is a diagram illustrating the construction of the Bayesian network in step S6 of the flowchart shown in FIG. 10.
A Bayesian network is a directed acyclic graph with random variables as nodes, and is a graph that has a conditional probability table between parent nodes and child nodes.
FIG. 11 shows four nodes represented by ellipses and four arrows (edges).
Nodes 301, 302, 311, and 312 are "normal + other", "gray mold", "10.0<first sensor value<20.0", and "250.0<nth sensor value<300.0", respectively. ”, and represents a random variable whose value is determined probabilistically. The node at the destination of the arrow is the child node, and the node at the base of the arrow is the parent node.

If event A is defined as "gray mold", event ^AC means "gray mold" is not present. Furthermore, if event B is defined as "250.0<nth sensor value<300.0", event BC indicates that "250.0<nth sensor value<300.0" is not satisfied.
Here, it is assumed that the number of training data for constructing a Bayesian network is 1000 items. Of these, 200 data items were determined to be "gray mold" and 800 data items were determined to be "normal or other". Also, among the 200 data items determined to be "gray mold", 6 data items were "10.0 < 1st sensor value <20.0", and 6 data items were "250.0 < nth sensor value < 300. There were 195 data items with "0". Of the 800 data items determined to be "normal or other," 106 data items were "10.0<1st sensor value<20.0" and 106 data items were "250.0<nth sensor value<300.0". ” There were 8 data items.

同様にして、「灰色カビ病」かつ「２５０．０＜第ｎセンサ値＜３００．０」である同時確率は、以下のように算出できる。

また、「灰色カビ病」だが「２５０．０＜第ｎセンサ値＜３００．０」でない同時確率は、以下のように算出できる。

また、「灰色カビ病」でなく「２５０．０＜第ｎセンサ値＜３００．０」でない同時確率は、以下のように算出できる。

このとき、「２５０．０＜第ｎセンサ値＜３００．０」である確率は以下のように算出できる。

従って、「２５０．０＜第ｎセンサ値＜３００．０」である場合に「灰色カビ病」である確率は、ベイズの定理により、以下のように算出できる。

これは、図１２において、ノード３１２からノード３０２に向かう破線の矢印に対応する条件付き確率が０．９６と算出されたことを示している。 The conditional probability P(B|A) that a crop with "gray mold disease" is "250.0<nth sensor value<300.0" is expressed as the ratio of the number of cases (frequency) of each of the above data. It is calculated as follows.

Similarly, the joint probability of "gray mold" and "250.0<nth sensor value<300.0" can be calculated as follows.

Further, the joint probability of "gray mold" but not "250.0<nth sensor value<300.0" can be calculated as follows.

Moreover, the joint probability that it is not "gray mold" and "250.0<nth sensor value<300.0" can be calculated as follows.

At this time, the probability that "250.0<nth sensor value<300.0" can be calculated as follows.

Therefore, when "250.0<nth sensor value<300.0", the probability of "gray mold" can be calculated as follows using Bayes' theorem.

This indicates that the conditional probability corresponding to the dashed arrow pointing from node 312 to node 302 in FIG. 12 is calculated to be 0.96.

As described above, when using training data, events indicating the growth state of agricultural products, which are judgment results, are set as parent nodes 301 and 302, and data or features obtained by discretizing the training data are set as child nodes 311 and 312, the parent nodes A Bayesian network 300 can be constructed by creating a conditional probability table between nodes and child nodes.
A known method can be used to construct a Bayesian network. For example, structural learning algorithms such as Greedy Search algorithm, Stingy Search algorithm, exhaustive search method, etc. can be used to construct a Bayesian network.

Next, when constructing a Bayesian network, conditions to be applied to the Bayesian network are selected (step S7). Regarding the selection of conditions to be applied to the Bayesian network, the Bayesian network adjustment unit 11f of the agricultural product growth state observation device 10 displays necessary information on the growth state display device 30, and allows the user to select the conditions to be applied to the Bayesian network. It may also be accepted via an input device.

In FIG. 12, only the first sensor value and the nth sensor value are explained as an example, but other sensor data such as the second sensor value, the first public data to the mth public data, and the first image recognition feature amount A Bayesian network 300 can be constructed including the k-th image recognition features. The conditions to be applied to the Bayesian network 300, that is, which data to incorporate as nodes can be selected as appropriate.

At this time, in FIG. 12, the nth sensor 50c is installed in the cultivation facility 200 and the sensor data is acquired to construct the Bayesian network 300, but the first sensor 50c is not actually installed. Even if the nth sensor data does not exist, if there is known knowledge regarding the relationship between the sensor data of the nth sensor 50c and the occurrence of gray mold, the probability etc. corresponding to the arrow 333 can be manually set. can be incorporated into the Bayesian network 300. This corresponds to steps S8 and S7 described later in the flowchart of FIG. 10, and corresponds to a case where there is known know-how and conditions to be applied are added.
In agriculture, where there is a lot of know-how accumulated over many years, the knowledge of experts can be incorporated into the Bayesian network 300. Furthermore, even if a sensor is added during the learning of the Bayesian network 300, it can be incorporated into the Bayesian network 300 as long as there is knowledge of how much it is related to the growth of agricultural products.

If there is known know-how regarding the growth of agricultural products (Yes in step S8), the process returns to step S7, and conditions to be applied to the Bayesian network are added or omitted based on the known know-how. Specifically, at least one of the nodes, arrows, and conditional probabilities (probability values) that constitute the Bayesian network is changed. The Bayesian network adjustment unit 11f of the crop growth condition observation device 10 inquires of the user through the growth condition display device 30 whether there is any known know-how regarding the growth of crops, and if the user answers that there is, the above-mentioned know-how is determined. A transition may be made to a screen that accepts selection of conditions to be applied to a Bayesian network such as the one described above. Further, in particular, the case where there is known know-how is accepted as a selection of conditions to be applied to the Bayesian network without any particular distinction, and the user is inquired through the growth status display device 30 whether or not to complete the selection of the conditions, and the selection is completed. If this is selected, it may be treated as if there is no known know-how.
If there is no known know-how regarding the growth of agricultural products, the generation of the Bayesian network is completed, the generated Bayesian network is stored in the Bayesian network storage unit 12, and the process proceeds to the stage of using the Bayesian network from step S9 onwards.

<Growth status estimation using Bayesian network>
Below, a procedure for estimating and predicting the growth state of agricultural products using the Bayesian network generated through the processing up to step S8 will be described. Here, the procedure of estimating and predicting the growth state of agricultural products using a Bayesian network corresponds to the method for estimating the growth state of agricultural products of the present invention.
First, data or feature amounts are collected from the sensor data of the installed sensor 50, the public data acquired from the public data acquisition unit 60, and the image data photographed by the camera 40, and the sensor data TSDB 20b, the public data TSDB 20c, and the image Each is stored in the data TSDB 20e (step S9). Here, sensor data, public data, and image data are acquired by the data acquisition unit 13a of the agricultural product growth state observation device 10, and if necessary, the raw data is subjected to feature calculation processing in the feature calculation unit 13b. You may calculate the feature amount by executing In this way, the acquired sensor data, etc. or feature quantities are stored in the sensor data TSDB 20b, public data TSDB, and image data TSDB 20e. Here, the data acquired by the camera 40, the sensor 50, and the public data acquisition unit 60 is data related to the crop 100 currently cultivated in the cultivation facility, and the crop 100 corresponds to the target crop of the present invention, and the feature amount The calculation unit 13b corresponds to the target agricultural product feature amount acquisition unit of the present invention. Further, the sensor data TSDB 20b, the public data TSDB, and the image data TSDB 20e correspond to the target agricultural product feature quantity storage unit of the present invention.

Next, the collected data or feature amounts are input to the Bayesian network (step S10). Here, in response to the user instructing the crop growth state observation device 10 via the input device to estimate the growth state of the farm products, the acquisition of data to be input into the Bayesian network is performed. The data acquisition unit 13c may acquire necessary data from the sensor data TSDB 20b, the public data TSDB 20c, and the image data TSDB 20e, and input the data to the Bayesian network. In addition, in order to obtain a desired estimation result of the growth state, the user specifies the range of data to be input to the Bayesian network via the input device by data acquisition date and time, etc., and the input data acquisition unit 13c responds to this instruction. Accordingly, corresponding data may be acquired from the sensor data TSDB or the like.

Next, the growth state of the crops is estimated from the probability calculated using the Bayesian network (step S11). Based on the input data, etc., the probability calculation unit 13d calculates the probability using a Bayesian network and estimates the growth state of the crops. The probability calculation section 13d corresponds to the probability calculation section of the present invention.

FIG. 13 is a diagram illustrating estimation of the growth state of agricultural products using a Bayesian network in step 10 and step S11 of the flowchart shown in FIG. 10. A Bayesian network 400 shown in FIG. 12 is a Bayesian network generated by the Bayesian network generation unit 11.

In the Bayesian network 400, each parameter of the first sensor value to the nth sensor value, the first public data value to the mth public data value, and the first image recognition feature amount to the kth image recognition feature amount is calculated using a conditional probability table. From the values, it is possible to determine the estimated probability of each event being "normal", "Botrytis", and "calcium deficiency". In FIG. 12, these are "10.0<first sensor value<20.0", "250.0<nth sensor value<300.0", and "700.0<first public data value<750.0". 0'', ``1000.0<first image recognition feature value<1100.0'' from nodes 411, 412, 413, and 414 representing parameter values of ``normal'', ``gray mold'', and ``calcium deficiency''. The numerical values attached to the arrows pointing to the respective nodes 401, 402, and 403 are the probabilities of each event based on each parameter value. In FIG. 12, cases where the probability exceeds 0.5 are indicated by thick dashed arrows.

Inputting data or feature amounts collected from sensor data, public data, and image data into the Bayesian network 400 described above, and estimating the growth state of crops such as normal, botrytis, calcium deficiency, etc. based on the calculated probability. (corresponds to steps S9 to S11 in the flowchart shown in FIG. 10).

Furthermore, in the estimation using the Bayesian network 400 described above, the growth state of agricultural products can be estimated not only when all data is acquired and input by the sensor 50 etc. but also when some data cannot be acquired. . For example, a case will be described in which sensor data cannot be acquired because the first sensor 50a is not installed or is installed but is malfunctioning. FIG. 14 shows the Bayesian network 400 shown in FIG. 13 again. Here, although the value of the first sensor 50a is not given as input data, the probability of an arrow with the value of the first sensor 50a as the parent node and the growth condition of crops such as calcium deficiency as the child node is expressed as follows: This does not mean that it cannot be used to estimate the growth state. As shown by the arrow 451 in FIG. If the value of another parameter such as "sensor value <300.0" is obtained as input data, the value of the first sensor 50a will be "10.0 < first sensor value < 20. 0'' can be calculated as 0.81 by belief propagation. Thereby, even if the value of the first sensor 50a is not obtained, it is possible to calculate the probability that the growth state of the agricultural product is estimated to be calcium deficient.

By using a Bayesian network in this way, it is possible to predict growth conditions even in situations where not all data is available as input data, so predictions can be made even if some data cannot be used due to noise due to external disturbances. can do. In outdoor crop cultivation environments, there are many disturbances, and input data for prediction may not be obtained appropriately. Even in such cases, it is possible to predict crop growth conditions by using a Bayesian network. Can be done properly.

Further, a screen showing the graph structure of the generated Bayesian network 400 may be generated by the output information generation unit 13e and displayed on the growth state display device 30. FIG. 15 shows an example of a Bayesian network display screen 500. Since the graph structure of the Bayesian network display screen 500 shown in FIG. 15 is the same as the Bayesian network 400 shown in FIG. 13, detailed explanation will be omitted. As shown in Figure 15, arrows whose conditional probabilities are less than a threshold (for example, 0.5) are indicated by thin broken lines, and arrows whose conditional probabilities are greater than or equal to the threshold are indicated by thick solid lines. Items with a high probability are indicated by a thicker solid line. In this way, the direction of the causal relationship is displayed by the direction of the arrow between the nodes, and the strength of the causal relationship (the size of the conditional probability) is visually They are displayed so that they can be clearly distinguished. In Figure 15, the strength of the causal relationship is displayed by varying the type and thickness of the arrow, but the display format can be changed depending on the strength of the causal relationship, and the arrow with the strongest causal relationship is displayed in red. The arrows may be displayed in different colors, such as . Here, the Bayesian network display screen 500 corresponds to the graph display screen of the present invention, and the output information generation section 13e corresponds to the screen generation section of the present invention.

When changing the type and thickness of the arrows depending on the strength of the causal relationship on the Bayesian network display screen 500, if the conditional probability is greater than or equal to a threshold value (for example, 0.5), the arrow is displayed and the conditional probability is Arrows below the threshold (arrows indicated by broken lines in FIG. 15) may not be displayed. In this way, the strength of the causal relationship between events can be visualized more clearly.

A Bayesian network models the probabilistic causal relationship of each variable set as a node using a network structure. Be able to identify the root cause of a condition (symptom). Furthermore, by displaying and visualizing the constructed Bayesian network 400 as the Bayesian network display screen 500, the user can understand the relationship structure of factors.

The Bayesian network that is generated is not limited to the Bayesian network 400 shown in FIG. 13. For example, as shown in FIG. 16, a Bayesian network 600 can be constructed by expanding the range of the network. Components common to the Bayesian network 400 are designated by the same reference numerals, and description thereof will be omitted. Here, the range of the network is expanded by considering the values of each parameter such as sensor values as events. In FIG. 16, an arrow 461 points from a node 411 representing "10.0<first sensor value<20.0" to a node 412 representing "250.0<nth sensor value<300.0", and "10.0 An arrow 462 from a node 411 representing <first sensor value <20.0 to a node 413 representing "700.0<first public data value <750.0", "10.0<first sensor value <20" From node 411 representing ".0" to node 412 representing "250.0<nth sensor value<300.0", node 413 representing "700.0<first public data value<750.0", and node 413 representing "1000.0". 0<first image recognition feature value<1100.0", and go to a node 401 representing "normal", a node 402 representing "gray mold", and a node 403 representing "calcium deficiency", respectively. Arrows 463, 464, and 465 have been added. The number of cases where "10.0<first sensor value<20.0" shown in node 41 is 782 out of 1000, the conditional probability corresponding to arrow 461 is 203/782, and the condition corresponding to arrow 462 The winning probability is calculated as 586/782. From "10.0<first sensor value<20.0" shown in node 41 to "250.0<nth sensor value<" from node 411 representing "10.0<first sensor value<20.0" Of the 203 items corresponding to the arrow pointing to the node 412 representing "300.0", there are 164 items corresponding to the arrow passing through the node 413 representing "700.0<first public data value <750.0". Among them, there are 113 cases that match the arrow passing through the node 414 representing "1000.0<first image recognition feature value<1100.0", and furthermore, the number of cases that match the arrow passing through the node 414 representing "normal", and the node 401 representing "normal", "gray mold disease". The numbers corresponding to the arrows pointing to the node 402 representing "calcium deficiency" and the node 403 representing "calcium deficiency" are 13, 30, and 78, respectively.

For such a Bayesian network 600, input 17.5 as the first sensor value, 283.1 as the nth sensor value, 726.9 as the first public data value, and 1021.4 as the first image recognition feature amount. (corresponding to step 10 in the flowchart of FIG. 10), the predicted probability that the crop has calcium deficiency is 78/113, the predicted probability that the crop has gray mold is 30/113, and the predicted probability that the crop is normal is 13/113. (corresponds to step S11 in the flowchart of FIG. 10). Therefore, the estimated growth status of crops is that calcium deficiency is the first, gray mold disease is the second, and normal is the third.

Next, each data and the estimated result are stored in the determination result TSDB 20f (step S12). The estimated result is stored in the determination result TSDB 20f by the probability calculation unit 13d in association with the data input to the Bayesian network.

Next, the output information generation unit 13e of the agricultural product growth state observation device 10 generates screen information that provides information on symptoms estimated from the pest and physiological disorder DB 20d in a ranking format, and displays it on the growth state display device 30 (step S13). Based on the calculated probabilities, the output information generation unit 13e of the crop growth state observation device 10 generates a growth state ranking screen that displays the growth state of the crops in a ranking format according to the probability of occurrence, and displays the growth state ranking screen in the growth state display device 30. to be displayed. The growth state ranking screen may also display information regarding the growth state stored in the pest/disease/physiological disorder DB 20d.

FIG. 17 shows an occurrence probability ranking screen 700 that is an example of a growth state ranking screen. On the occurrence probability ranking screen 700, the growth status of agricultural products estimated by the Bayesian network 400 or the like is displayed in a ranking format. Here, the first place is gray mold, the second place is calcium deficiency, and the third place is powdery mildew. The occurrence probability ranking screen 700 each shows a rank display 701 of 1st, 2nd, and 3rd place, an occurrence probability display 702 expressed as a percentage, a growth state name 703 of gray mold, a summary information 704 of the growth state, and a growth state summary information 704. It is composed of a photographic image 705. Here, the occurrence probability ranking screen 700 corresponds to a growth state display screen and a growth state ranking display screen of the present invention. Furthermore, although a screen is described here that displays three growth states in descending order of probability of occurrence, the output information generation unit 13e screen generation unit creates a screen that displays only the growth state with the highest probability of occurrence. It may be generated and displayed on the growth state display device 30. At this time, the screen that displays only the growth state with the highest probability of occurrence corresponds to the growth state display screen of the present invention.

The occurrence probability is displayed as a numerical value estimated by inputting sensor data etc. to the Bayesian network 400 or the like.
Further, the summary information 707 and the photographic image 705 are acquired from the pest/disease/physiological disorder information DB 20d and displayed.

In this way, by using a Bayesian network, multiple targets can be predicted simultaneously. In this way, there is no distinction between objective variables and explanatory variables, and multiple predictions are possible instead of just one prediction target, so predictions of crop symptoms can be provided in a ranking format. By targeting multiple similar symptoms, it is possible to cover all possible symptoms.

<Additional note 1>
a target crop feature acquisition unit (13a, 13b) that acquires a target crop feature amount that is the feature amount regarding the target crop that is currently being cultivated;
a target crop feature amount storage unit (20b, 20c, 20e) that stores the target crop feature amount;
a probability calculation unit (13d) that calculates the probability of the growth state of the target crop using a Bayesian network model (400, 600) based on the target crop feature;
a screen generation unit (13e) that generates a growth state display screen that displays the estimated growth state according to the calculated ranking of the probabilities;
A display section (30);
A crop growth state estimation device (1) characterized by comprising:

1: Crop growth state observation system 13a: Data acquisition section 13b: Feature amount calculation section 13d: Probability calculation section 13e: Output information generation section

Claims (6)

a target crop feature amount acquisition unit that acquires a target crop feature amount that is a feature amount regarding the target crop that is currently cultivated;
a target crop feature amount storage unit that stores the target crop feature amount;
a probability calculation unit that calculates the probability of the growth state of the target crop using a Bayesian network model based on the target crop feature;
a screen generation unit that generates a growth state display screen that displays the estimated growth state according to the calculated probability;
A display section;
A crop growth state estimation device characterized by comprising: The screen generation unit includes:
The crop growth state estimating device according to claim 1, wherein a growth state ranking display screen is generated that displays the estimated growth state in association with the calculated probability ranking. A teacher data storage unit that stores teacher data in which a feature amount acquired from growth-related information related to the growth of the agricultural product is associated with growth state information indicating the growth state of the agricultural product targeted by the growth-related information. and,
a Bayesian network model construction unit that constructs a Bayesian network model using the teacher data stored in the teacher data storage unit;
It is equipped with a Bayesian network model generation unit having
The crop growth state estimating device according to claim 1 or 2, wherein the probability calculation unit calculates the probability of the growth state using the Bayesian network model generated by the Bayesian network model generation unit. The Bayesian network model generation unit includes:
The agricultural product according to claim 3, further comprising a Bayesian network model adjustment unit that changes at least one of nodes, edges, and probability values that constitute the Bayesian network model constructed by the Bayesian network model construction unit. Growth state estimation device. The screen generation unit includes:
2. The crop growth state estimation device according to claim 1, wherein the device generates a graph display screen that displays a graph structure of the Bayesian network model. a step of acquiring a target crop feature amount that is a feature amount related to the target crop that is currently being cultivated;
a step of storing the feature amount of the target crop; a step of calculating a probability of the growth state of the target crop using a Bayesian network model based on the feature amount of the target crop;
generating a growth state display screen that displays the estimated growth state according to the calculated probability;
Displaying the growth status display screen;
A method for estimating the growth state of agricultural crops, comprising:

Images