Soil diagnostic device, learning device, and soil diagnostic system

Abstract

To provide a soil diagnostic device capable of accurately obtain soil analysis results, a learning device, and a soil diagnostic system.SOLUTION: A soil diagnostic device 1 comprises: a property data acquisition unit 11 which acquires property data relating to physical and / or chemical properties of soil in a field; a farming data acquisition unit 12 which acquires farming data related to agricultural operations in the field; and a generation unit 13 which generates soil diagnosis result data by inputting the property data and the farming data into a pre-generated learning model. The learning model is trained using training input data including training property data and training farming data which were used for soil diagnosis by a plurality of diagnosticians as explanatory variables, and training diagnosis result data generated by the plurality of diagnosticians on the basis of other training input data similar to the training input data as objective variables.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a soil diagnostic device, a learning device, and a soil diagnostic system. [Background technology]

[0002] Regarding technology for obtaining soil diagnostic results, a method for determining a fertilization design is known, as described in Patent Document 1. In this method, a sample collected from the soil is chemically analyzed to obtain chemical data, and its physical properties are analyzed to obtain physical data. In addition, this method calculates the amount of fertilizer to be applied based on the chemical data and the physical data, and determines the fertilization design for the soil. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-103889 Summary of the Invention [Problem to be solved by the invention]

[0004] In the past, soil diagnosis was sometimes performed by a single qualified soil doctor. In this case, the soil diagnosis results were generated based on the knowledge of a single soil doctor, and there was a possibility that the soil diagnosis results could not be obtained accurately due to differences in the level of knowledge among soil doctors.

[0005] The present invention has been made in view of the above circumstances, and has as its object to obtain soil diagnostic results with high accuracy. [Means for solving the problem]

[0006] A soil diagnostic device according to one embodiment of the present invention comprises a property data acquisition unit that acquires property data relating to the physical or chemical properties of soil in a field, a farming data acquisition unit that acquires farming data relating to farming operations in the field, and a generation unit that inputs the property data and farming data into a pre-generated learning model to generate soil diagnostic result data, and the learning model is trained using learning input data including learning property data and learning farming data used in soil diagnosis by multiple diagnosticians as explanatory variables, and learning diagnostic result data generated by multiple diagnosticians based on other learning input data similar to the learning input data as a dependent variable.

[0007] In one aspect of the present invention, a soil diagnostic device generates soil diagnostic result data by inputting property data and farming data into a learning model. This learning model is trained using learning input data, including learning property data and learning farming data used by multiple diagnosticians, as explanatory variables, and learning diagnostic result data generated by multiple diagnosticians based on other learning input data similar to the learning input data as a response variable. By inputting the property data and farming data into the learning model, diagnostic result data based on the knowledge of multiple diagnosticians can be output. Therefore, soil diagnostic results can be obtained with high accuracy.

[0008] The soil diagnostic device may further include a notification unit that notifies anomaly information indicating soil abnormalities based on the diagnostic result data generated by the generation unit. For example, if an abnormality is found in the soil compared to normal years, the soil abnormality can be notified to the diagnostician. This allows the cause of the soil abnormality to be quickly identified.

[0009] In the soil diagnostic device, the property data may include at least one of the measurement results of the chemical composition of the soil and the measurement results of the three-phase distribution of the soil. In this case, the scientific measurement results are input to the learning model as property data, so that diagnostic result data based on objective data can be generated.

[0010] In the above soil diagnostic device, the farming data may include at least one of the results of agricultural practice of the previous crop in the field and the cultivation plan of the next crop in the field. In this case, for example, the state of the field from the viewpoint of the farmer of the field is input as the farming data, so that diagnostic result data can be generated based on not only scientific data but also the state of the field from the viewpoint of a person.

[0011] A learning device according to another aspect of the present invention includes a learning data acquisition unit that acquires learning input data, including learning property data related to the physical or chemical properties of soil in a field and learning farming data related to farming operations in the field, used in soil diagnosis by multiple assessors, and learning diagnosis result data generated by the multiple assessors based on other learning input data similar to the learning input data, and a model generation unit that generates a learning model by learning using the learning input data as explanatory variables and the learning diagnosis result data as a target variable. This learning device achieves the same effects as the soil diagnostic device described above.

[0012] The learning device may further include a determination unit that determines the learning diagnostic result data for a target field based on reference data including property data and farming data for a plurality of preset reference fields and learning input data for a target field for which the learning diagnostic result data is to be determined. For example, a good field, which is a field that has achieved good crop cultivation results, can be set as the reference field. In this case, the diagnostic result data for the good field can be determined as the learning diagnostic result data for the target field, and a learning model can be generated using the diagnostic result data for the good field as the objective variable. This makes it possible to generate a learning model that can generate diagnostic result data that brings the condition of the target field closer to that of a good field when the property data and farming data are input. This allows for more accurate soil diagnostic results to be obtained.

[0013] In the learning device, the determination unit may identify a reference field among multiple reference fields whose state is closest to the state of the target field based on the reference data and the learning input data for the target field, and determine the diagnostic result data of the identified reference field as the diagnostic result data for learning of the target field. For example, if a good field is set as the reference field, a good field whose state is closest to the state of the target field can be identified. Because the state of the identified good field is close to the state of the target field, it is considered that the amount of work required to bring the state of the target field closer to the state of the good field is small. Furthermore, a learning model can be generated using the diagnostic result data of the identified good field as the objective variable, and a learning model can be generated that can generate diagnostic result data to bring the state of the target field closer to the state of the good field that is closest to the state of the target field when property data and farming data are input. Therefore, it is possible to obtain a diagnosis result that easily brings the state of the farm field closer to that of a good farm field.

[0014] In the above learning device, the determination unit may express each of the learning input data and the reference data of the subject field as coordinate points in a two-dimensional coordinate system by dimensionally compressing the learning input data and the reference data of the subject field, and identify the reference field whose state is closest to that of the subject field based on the distance between the coordinate points of the learning input data and the coordinate points of the reference data of the subject field. In this case, because the learning input data and the reference data of the subject field are dimensionally compressed, the amount of calculation required to determine the reference field whose state is closest to that of the subject field can be reduced.

[0015] In the above-described learning device, the learning data acquisition unit may acquire learning input data and learning diagnosis result data for each soil type, and the model generation unit may generate a learning model for each soil type. For example, a fertilization design for supplying nutrients necessary for crop cultivation may differ for each soil type. Since a learning model is generated for each soil type, it is possible to generate a learning model that can generate diagnosis result data according to the soil type.

[0016] In the learning device, the learning input data may include first data, which is learning input data used by the assessor to diagnose the soil at a certain point in time, and second data, which is learning input data used by the assessor to diagnose the soil at a point in time prior to the first point in time. The model generation unit may generate the learning model by weighting the first data more heavily than the second data for use in learning. For example, due to fluctuations in weather conditions, the first data used to diagnose the soil at a certain point in time may be considered to better reflect the state of the field than the second data used to diagnose the soil at a point in time prior to the first point in time. By weighting the first data more heavily than the second data for use in learning, learning is performed with an emphasis on data closer to the time at which the learning model is generated. As a result, the learning model can be generated taking into account time-series changes in the state of the field, thereby obtaining more accurate soil diagnosis results.

[0017] The learning device may further include a receiving unit that receives a selection of excluded data, including learning property data or learning farming data that is not used in training the learning model. The model generation unit may generate a learning model using the learning input data that does not include the excluded data received by the receiving unit as explanatory variables. For example, data including learning property data or farming data that is considered inappropriate for use in training the learning model due to delays in farm work, farming errors, etc. may be selected as excluded data. In this case, the accuracy of learning can be improved by generating a learning model using learning input data that does not include the excluded data as explanatory variables. Therefore, more accurate soil diagnostic results can be obtained.

[0018] A soil diagnostic system according to yet another aspect of the present invention includes the soil diagnostic device and the learning device described above. This soil diagnostic system has the same effects as the soil diagnostic device and the learning device described above. [Effects of the Invention]

[0019] The present invention aims to obtain soil diagnostic results with high accuracy. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 1 is a schematic diagram showing an application example of a soil diagnostic system according to an embodiment of the present invention. [Figure 2] FIG. 10 is a diagram illustrating an example of diagnosis result data. [Figure 3] FIG. 10 is a diagram showing an example of chemical property data. [Figure 4] FIG. 10 is a diagram illustrating an example of physical property data. [Figure 5] FIG. 10 is a diagram showing an example of farming data. [Figure 6] 1 is a block diagram showing a configuration of a soil diagnostic system according to an embodiment of the present invention. [Figure 7] 3 is a flowchart showing an operation method of the soil diagnostic device according to the present embodiment. [Figure 8] 10 is a flowchart showing an operation method of the learning device according to the present embodiment. [Figure 9] (a) is a diagram showing an example of one step of determining diagnostic result data for learning. (b) is a diagram showing an example of the step following the step shown in (a). (c) is a diagram showing an example of the step following the step shown in (b). DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In each drawing, the same or corresponding parts are designated by the same reference numerals, and redundant explanations will be omitted.

[0022] FIG. 1 is a schematic diagram showing an application example of a soil diagnostic system 10 according to this embodiment. FIG. 2 is a diagram showing an example of diagnostic result data. The soil diagnostic system 10 is a system for diagnosing soil in a farm field. A "farm field" is agricultural land for cultivating crops. A farm field includes, for example, fields and paddy fields. "Diagnosing the soil in a farm field" includes outputting diagnostic result data of the soil. The soil diagnostic system 10 outputs the diagnostic result data to, for example, a user of the soil diagnostic system 10.

[0023] As shown in FIG. 2, the diagnostic result data includes, for example, the current state of the soil and remedial measures for improving the soil condition. The current state of the soil includes, for example, at least one of the physical properties and chemical properties of the soil. The physical properties of the soil include, for example, measurement results of the three-phase distribution of the soil. The chemical properties of the soil include, for example, measurement results of the chemical composition of the soil. The remedial measures include, for example, a fertilization design for improving the chemical composition of the soil and the details of work to be performed on the field. However, the contents of the diagnostic result data are not limited to those described above and can be changed as appropriate.

[0024] For example, a user of the soil diagnostic system 10 performs work on a field based on the diagnostic result data output by the soil diagnostic system 10. The user improves the soil by applying fertilization to the soil according to the fertilization design included in the diagnostic result data. Before the user performs work based on the diagnostic result data, for example, a person who checks the diagnostic result data may confirm the validity of the diagnostic result data. The person who checks the diagnostic result data may be, for example, a person who is qualified as a soil doctor (hereinafter referred to as a "soil doctor").

[0025] The soil diagnostic system 10 outputs diagnostic result data based on the property data and farming data. The property data is data relating to the physical or chemical properties of the soil in the field. The property data is generated, for example, by a soil analyst P1. The analyst P1 is, for example, a person with physical or chemical knowledge of soil. The property data includes, for example, chemical property data relating to the chemical properties of the soil and physical property data relating to the physical properties of the soil.

[0026] Fig. 3 is a diagram showing an example of chemical property data. The chemical property data includes, for example, identification information for uniquely identifying a sample for analyzing chemical properties ("sample information" in the example of Fig. 3). The chemical property data includes, for example, the history of the field ("field history (previous year)" in the example of Fig. 3). The history of the field includes, for example, at least one of the details of the previous crop in the field, the application history of organic matter, and the application history of green manure.

[0027] The "previous crop" refers to, for example, a crop cultivated in a field at a time before the diagnostic result data is generated by the soil diagnostic system 10. Note that the "succeeding crop" described below refers to a crop cultivated in a field after the previous crop.

[0028] The chemical property data includes, for example, measurement results of the chemical composition of the soil ("chemical analysis results" in the example of Figure 3). The measurement results of the chemical composition of the soil include, for example, multiple analysis items including numerical values ​​indicating the chemical properties of the soil. The multiple analysis items are classified, for example, into general items, trace elements, nitrogen, soil properties, and other analysis items. The general items are items that indicate the general properties of the soil. The trace elements are items that indicate trace components contained in the soil. Nitrogen is an item that indicates components that contain nitrogen (for example, nitrogen compounds). The soil properties are, for example, items that indicate soil properties other than the general items. The other analysis items are, for example, items that indicate soil properties other than the general items, trace elements, nitrogen, and soil properties.

[0029] The chemical property data includes, for example, analytical values ​​(measured values) for each of a plurality of analytical items and reference values ​​that serve as the reference for the analytical values. The reference values ​​have, for example, a predetermined numerical range. The reference values ​​include, for example, a reference lower limit value that is the lower limit of the numerical range and a reference upper limit value that is the upper limit of the numerical range. The chemical property data includes, for example, nutrient status based on the analytical values ​​and reference values ​​for each of a plurality of analytical items. For example, if the analytical value for one analytical item is smaller than the reference value, the nutrient status can be said to be low; conversely, if the analytical value is larger than the reference value, the nutrient status can be said to be high. The nutrient status can be expressed in five levels, for example, "low," "slightly low," "reference value," "slightly high," and "high."

[0030] The chemical property data includes, for example, a radar chart showing analytical values ​​and reference values ​​for each representative value indicating the properties of the soil. In the example of Fig. 3, the radar chart includes the soil pH, lime content, magnesium content, potassium content, phosphate content, and hydrothermal nitrogen (hot water extractable nitrogen) content.

[0031] The chemical property data includes, for example, the required amount of material to improve the condition of the soil ("required amount of soil improvement material" in the example of FIG. 3). The required amount of material is calculated, for example, based on the analytical values ​​and the reference values. Specifically, the required amount of material is calculated so that the numerical values ​​for each of the multiple analytical items fall within the numerical range of the reference values.

[0032] The chemical property data includes, for example, the result of fertilization design ("Fertilization Design Result" in the example of FIG. 3). The result of fertilization design is calculated, for example, based on the analysis values ​​and the reference values. The result of fertilization design is, for example, the amount of fertilizer calculated so that the values ​​for each of the multiple analysis items fall within the range of the reference values.

[0033] FIG. 4 is a diagram showing an example of physical property data. The physical property data includes, for example, measurement results of the three-phase distribution of soil. Soil is a substance composed of, for example, a solid phase, a liquid phase, and a gas phase. The three-phase distribution is the volume ratio of each of the solid phase, the liquid phase, and the gas phase. The three-phase distribution is one of the indicators of soil fertility and can be used for soil diagnosis. The three-phase distribution may be measured, for example, using a known penetration-type hardness tester.

[0034] The physical property data includes, for example, a three-phase distribution for each soil type, a three-phase distribution for each soil layer type, and a solid fraction, a moisture fraction, and an air fraction.

[0035] The farming data is data related to farming of a field. The farming data is generated, for example, by a farmer P2 who farms the field. The farmer P2 is, for example, a person who works in the field. The farmer P2 is, for example, a farm worker.

[0036] Fig. 5 is a diagram showing an example of farming data. The farming data includes, for example, identification information for uniquely identifying a field. The farming data includes, for example, an overview of the field ("1. Overview of field" in the example of Fig. 5). The overview of the field includes, for example, at least one of drainage conditions, soil fertility, and cropping system.

[0037] The farming data includes, for example, the results of agricultural practices for the previous crop in the field. The results of agricultural practices for the previous crop include the details of the work that was carried out in the field for the previous crop. The results of agricultural practices for the previous crop include, for example, at least one of the cultivation method, the details of the previous crop (including at least one of the crop name and variety name), the cultivation period, and the yield (in the example of Figure 5, part of the content of "Cultivation period and yield of the previous crop and the planned successive crop"). The results of agricultural practices for the previous crop include, for example, at least one of the application history of compost, the application history of soil improvement materials, and the details of the fertilizer (including at least one of the fertilizer name, application amount, and application date) (in the example of Figure 5, "3. Application status of compost, soil improvement materials, and fertilizer for the previous crop"). The results of agricultural practices for the previous crop include, for example, at least one of the growth status of the previous crop, the rate of saleable crop, whether or not there are physiological disorders, whether or not there are diseases, whether or not there are insect damage, and problems (in the example of Figure 5, "4. Growth status of the previous crop, the rate of saleable crop, physiological disorders, pests and diseases, problems, etc.").

[0038] The farming data includes, for example, the cultivation plan for the subsequent crop in the field (in the example of Figure 5, part of the content of "2. Cultivation period and yield of the previous crop and the subsequent crop plan"). The cultivation plan for the subsequent crop includes, for example, at least one of the name of the subsequent crop, the name of the variety, the date of sowing, and the date of planting.

[0039] The above describes examples of property data and farming data, but the property data may be data related to the physical or chemical properties of soil. The farming data may be data related to farming of a field. The content of the property data and farming data can be changed as appropriate and is not limited to the content described above. In addition, in the above example, an example was described in which the property data included both the measurement results of the soil's chemical composition and the measurement results of the soil's three-phase distribution. However, the property data may include at least one of the measurement results of the soil's chemical composition and the measurement results of the soil's three-phase distribution. Furthermore, in the above example, an example was described in which the farming data included both the agricultural practice results of the previous crop in the field and the cultivation plan for the subsequent crop in the field, but the farming data may include at least one of the agricultural practice results of the previous crop and the cultivation plan for the subsequent crop.

[0040] Conventionally, soil diagnosis has sometimes been performed by a single person (hereinafter referred to as "diagnostician"). The diagnostician is, for example, a soil doctor. The diagnostician generates diagnostic result data using his or her own knowledge, for example, based on the property data and farming data described above. However, since the soil diagnostic result is generated depending on the knowledge of a single diagnostician, there is a possibility that the soil diagnostic result cannot be obtained accurately due to reasons such as differences in the level of knowledge of the diagnosticians. The soil diagnostic system 10 according to this embodiment has the configuration described below in order to solve the above problems.

[0041] FIG. 6 is a block diagram showing the configuration of a soil diagnostic system 10 according to this embodiment. The soil diagnostic system 10 includes a soil diagnostic device 1, a field database D1, a diagnostic result database D2, a learning device 2, and a user terminal 3. In the example of FIG. 6, only one user terminal 3 is shown, but the soil diagnostic system 10 may include multiple user terminals 3. The soil diagnostic device 1, the field database D1, the diagnostic result database D2, the learning device 2, and the user terminal 3 are capable of communicating with each other, for example, via a network N. The configuration of the network N is not particularly limited. The network N may include, for example, the Internet or an intranet.

[0042] The soil diagnostic device 1 generates diagnostic result data based on property data and farming data. The soil diagnostic device 1 inputs the property data and farming data into a pre-generated learning model to generate diagnostic result data. The learning model is trained using learning input data (described later) as explanatory variables and learning diagnostic result data as objective variables. The soil diagnostic device 1 may store the pre-generated learning model.

[0043] The farm field database D1 stores learning input data. The learning input data is used as explanatory variables in learning the learning model. The learning input data includes learning property data and learning farming data used in soil diagnosis by multiple assessors. The learning input data is data used in soil diagnosis in the past. "Past" means, for example, a time point prior to the time when the diagnosis result data is generated by the soil diagnosis system 10.

[0044] The learning input data includes type information indicating the type of soil. The learning input data includes time point information indicating the time point when the learning input data was generated. The time point information includes, for example, the year, month, and day when the learning input data was generated. The learning input data includes first data, which is learning input data used by the assessor to diagnose the soil at a certain time point, and second data, which is learning input data used by the assessor to diagnose the soil at a time point prior to the certain time point. The farm field database D1 stores learning input data for each of a plurality of farm fields. The farm field database D1 may store learning input data transmitted from the user terminal 3, for example.

[0045] The field database D1 stores, for example, reference data. The reference data is data including property data and farming data of a plurality of preset reference fields. A reference field is, for example, a field where crops have been cultivated in the past and at least one of the crop yield and quality (including, for example, the rate of saleable products) was above a predetermined standard. For example, a field that has produced excellent crop cultivation results is selected as a reference field. The field database D1 stores reference data for each of a plurality of reference fields.

[0046] The diagnostic result database D2 stores diagnostic result data for learning. The diagnostic result data for learning is used as a target variable in learning of the learning model. The diagnostic result data for learning is generated by multiple diagnosers based on other learning input data similar to the learning input data. The diagnostic result database D2 stores diagnostic result data for learning for each of multiple fields. The diagnostic result database D2 may store, for example, diagnostic result data for learning transmitted from the user terminal 3. The diagnostic result database D2 stores diagnostic result data for learning linked to the learning input data stored in the field database D1.

[0047] The diagnostic result database D2 stores, for example, diagnostic result data of a reference field. The diagnostic result database D2 stores diagnostic result data for each of a plurality of reference fields. The diagnostic result database D2 stores diagnostic result data of reference fields linked to the reference data stored in the field database D1.

[0048] In the above example, the farm field database D1 stores the learning input data, and the diagnostic result database D2 stores the learning diagnostic result data. However, the learning input data and the learning diagnostic result data may be stored in a single database. The farm field database D1 and the diagnostic result database D2 are stored, for example, in a device (e.g., a server) external to the soil diagnostic device 1 and the learning device 2. However, the farm field database D1 and the diagnostic result database D2 may also be stored in the learning device 2.

[0049] The learning device 2 generates a learning model based on the learning input data stored in the farm field database D1 and the learning diagnosis result data stored in the diagnosis result database D2. The learning device 2 generates the learning model by learning using the learning input data as explanatory variables and the learning diagnosis result data as objective variables. The learning device 2 may, for example, transmit the generated learning model to the soil diagnostic device 1.

[0050] The soil diagnostic device 1 and the learning device 2 are computers, and physically include memories such as RAM (Random Access Memory) and ROM (Read Only Memory), processors (arithmetic circuits) such as CPU (Central Processing Unit), communications interfaces, and storage units such as hard disks. The soil diagnostic device 1 and the learning device 2 function by executing programs stored in the memories on the CPU of the computer system. The soil diagnostic device 1 and the learning device 2 may include a microcomputer or FPGA (Field Programmable Gate Array).

[0051] The user terminal 3 is a terminal used by a user. The user terminal 3 is, for example, a smartphone, a tablet terminal, or a personal computer. The user terminal 3 accepts input of property data and farming data. The user terminal 3 transmits the accepted property data and farming data to the soil diagnostic device 1. The user terminal 3, for example, receives diagnostic result data from the soil diagnostic device 1. The user terminal 3, for example, outputs the received diagnostic result data. The user terminal 3, for example, outputs the diagnostic result data via an output unit such as a display.

[0052] The user terminal 3, for example, accepts input of learning input data and learning diagnostic result data. The user terminal 3, for example, transmits the accepted learning input data to the farm field database D1 via the network N. The user terminal 3, for example, transmits the accepted learning diagnostic result data to the diagnostic result database D2 via the network N.

[0053] The user terminal 3 receives anomaly information (described later) from the soil diagnostic device 1. For example, when the user terminal 3 receives the anomaly information, it outputs the anomaly information via an output unit. For example, the user terminal 3 accepts input of exclusion data (described later) by a user. The user terminal 3 transmits the accepted exclusion data to the learning device 2.

[0054] The soil diagnostic device 1 includes, as its functional configuration, a property data acquisition unit 11, a farming data acquisition unit 12, a generation unit 13, and a notification unit 14.

[0055] The property data acquisition unit 11 acquires property data. The property data acquisition unit 11 receives and acquires property data input via, for example, the user terminal 3. The farming data acquisition unit 12 acquires farming data. The farming data acquisition unit 12 receives and acquires farming data input via, for example, the user terminal 3.

[0056] The generation unit 13 inputs the property data acquired by the property data acquisition unit 11 and the farming data acquired by the farming data acquisition unit 12 into a pre-generated learning model to generate soil diagnosis result data. The generation unit 13 transmits the generated diagnosis result data to the user terminal 3, for example.

[0057] The notification unit 14 notifies abnormality information indicating an abnormality in the soil based on the diagnosis result data generated by the generation unit 13. The notification unit 14, for example, transmits the abnormality information to the user terminal 3. The notification unit 14 causes the user terminal 3 to notify the abnormality information.

[0058] The notification unit 14 determines whether an abnormality has occurred in the soil based on the diagnostic result data. The notification unit 14 may determine whether an abnormality has occurred in the soil using known means. For example, the notification unit 14 determines whether at least one of the three-phase distribution and chemical composition of the soil deviates from a predetermined standard by a predetermined amount or more. The standard may be set, for example, according to at least one of the three-phase distribution and chemical composition of the soil measured when a soil diagnosis was performed in the past. If it is determined that at least one of the three-phase distribution and chemical composition of the soil deviates from the standard by a predetermined amount or more, the notification unit 14 determines that an abnormality has occurred in the soil and notifies abnormality information. When chemical analysis is performed, for example, abnormality information (alert) may be issued if the diagnostic standard is periodically (for several days) outside the standard (for example, the pH of paddy rice is in the range of 5.5 to 6.5). If it is determined that at least one of the three-phase distribution and chemical composition of the soil does not deviate from the standard by a predetermined amount or more, the notification unit 14 determines that no abnormality has occurred in the soil and does not send abnormality information to the user terminal 3. For example, a countermeasure may be issued, and if the yield has decreased by a predetermined amount (for example, 10%) or more compared to last year due to a cause other than abnormal weather or sudden disease (entered in the yield field of the farming information for the following year), abnormality information (alert) may be issued as an unexpected abnormality.

[0059] The learning device 2 includes, as its functional configuration, a receiving unit 21, a determining unit 22, a learning data acquiring unit 23, and a model generating unit 24.

[0060] The reception unit 21 receives the selection of excluded data. The excluded data includes learning property data or learning farming data that is not used in learning the learning model. The excluded data is, for example, data that is considered inappropriate for use in learning the learning model. As the excluded data, for example, learning property data or learning farming data when delays in farm work, mistakes in farm work, etc. occur is selected.

[0061] The determination unit 22 determines the diagnosis result data for learning of the target field (described later) based on the reference data stored in the field database D1 and the learning input data for the target field (described later). The determination unit 22 determines the learning diagnosis result data to be used for learning the learning model, for example. The processing of the determination unit 22 will be described later.

[0062] The learning data acquisition unit 23 acquires learning input data stored in the farm field database D1 and learning diagnostic result data stored in the diagnostic result database D2. The learning data acquisition unit 23 receives and acquires learning input data from the farm field database D1. The learning data acquisition unit 23 receives and acquires learning diagnostic result data from the diagnostic result database D2. The learning data acquisition unit 23 acquires learning input data and learning diagnostic result data for each soil type. The learning data acquisition unit 23 acquires learning input data and learning diagnostic result data linked to the learning input data for each soil type based on type information included in the learning input data.

[0063] The model generation unit 24 generates a learning model by learning using the learning input data acquired by the learning data acquisition unit 23 as explanatory variables and the learning diagnosis result data acquired by the learning data acquisition unit 23 as objective variables. The model generation unit 24 generates a learning model for each type of soil. The model generation unit 24 identifies the generation order of the learning input data based on time point information included in the learning input data. The model generation unit 24 generates a learning model by weighting the first data more heavily than the second data based on the identified generation order and using it for learning. The model generation unit 24 generates a learning model using the learning input data that does not include the excluded data received by the reception unit 21 as explanatory variables.

[0064] The model generation unit 24 may generate the learning model by a known learning method. For example, the model generation unit 24 may generate the learning model by at least one of supervised learning, unsupervised learning, and reinforcement learning. Examples of supervised learning algorithms include support vector machines, logistic regression, random forests, decision trees, k-nearest neighbors, perceptrons, and neural networks. Examples of unsupervised learning algorithms include k-means, principal component analysis, and self-organizing maps. Examples of reinforcement learning algorithms include Q-learning, Monte Carlo methods, and SARSA.

[0065] Next, an operation method of the soil diagnostic device 1 according to this embodiment will be described. Fig. 7 is a flowchart showing the operation method of the soil diagnostic device 1 according to this embodiment. Before the soil diagnostic device 1 starts operation, the soil diagnostic device 1 stores a learning model generated in advance by the learning device 2.

[0066] First, the property data acquisition unit 11 acquires property data (step S1). In step S1, the user terminal 3, for example, accepts input of property data. The user terminal 3 transmits the accepted property data to the soil diagnostic apparatus 1 via the network N. The property data acquisition unit 11 acquires the received property data.

[0067] Next, the farming data acquisition unit 12 acquires the farming data (step S2). In step S2, the user terminal 3, for example, accepts input of farming data. The user terminal 3 transmits the accepted farming data to the soil diagnostic device 1 via the network N. The farming data acquisition unit 12 acquires the received farming data.

[0068] Next, the generation unit 13 inputs the property data acquired in step S1 and the farming data acquired in step S2 into a pre-generated learning model to generate soil diagnosis result data (step S3). In step S3, the generation unit 13 transmits the generated diagnosis result data to the user terminal 3. The user terminal 3 outputs the received diagnosis result data via the output unit.

[0069] Next, the notification unit 14 determines whether or not an abnormality has occurred in the soil based on the diagnosis result data generated in step S3 (step S4). If it is determined that no abnormality has occurred in the soil (step S4: NO), the soil diagnostic device 1 ends the series of operations. If it is determined that an abnormality has occurred in the soil (step S4: YES), the notification unit 14 notifies the abnormality information (step S5). In step S5, the notification unit 14 transmits the abnormality information to the user terminal 3. The user terminal 3 outputs the received abnormality information via the output unit. After the above processes, the soil diagnostic device 1 ends the series of operations.

[0070] Next, an operation method of the learning device 2 according to this embodiment will be described. FIG. 8 is a flowchart showing the operation method of the learning device 2 according to this embodiment. Before the learning device 2 starts operation, the field database D1 stores learning input data for each of a plurality of fields. The field database D1 also stores reference data for each of a plurality of reference fields. The diagnostic result database D2 stores learning diagnostic result data for each of a plurality of fields. The diagnostic result database D2 stores learning diagnostic result data linked to the learning input data stored in the field database D1. The diagnostic result database D2 stores reference field diagnostic result data for each of a plurality of reference fields. The diagnostic result database D2 stores diagnostic result data for the reference field linked to the reference data stored in the field database D1.

[0071] First, the receiving unit 21 receives the selection of exclusion data (step S11). In step S11, the user selects exclusion data from the learning input data stored in the field database D1. The user terminal 3 receives the input of exclusion data. The user terminal 3 transmits the received exclusion data to the learning device 2. The receiving unit 21 receives the received exclusion data. The receiving unit 21 modifies the field database D1 based on the received exclusion data. Specifically, the receiving unit 21 deletes, for example, at least one of the learning property data and the learning farming data, which are the exclusion data, from the field database D1. However, the receiving unit 21 may add a flag indicating that the data is exclusion data to at least one of the learning property data and the learning farming data, which are the exclusion data.

[0072] Next, the determination unit 22 determines the diagnostic result data for learning of the target field based on the reference data stored in the field database D1 and the learning input data for the target field (step S12). In step S12, the determination unit 22 corrects the diagnostic result database D2 based on the determined diagnostic result data for learning. Specifically, the determination unit 22 corrects the diagnostic result data for learning stored in the diagnostic result database D2 to the determined diagnostic result data for learning. Step S12 will be described in detail later.

[0073] Next, the learning data acquiring unit 23 acquires the learning input data stored in the farm field database D1 and the learning diagnostic result data stored in the diagnostic result database D2 (step S13). In step S13, the learning data acquiring unit 23 acquires, for each soil type, the learning input data and the learning diagnostic result data linked to the learning input data based on the type information. The learning data acquiring unit 23 acquires the learning input data that does not include the excluded data accepted in step S11. In step S12, if the determining unit 22 has corrected the learning diagnostic result data, the learning data acquiring unit 23 acquires the corrected learning diagnostic result data.

[0074] Next, the model generation unit 24 generates a learning model by learning using the learning input data acquired in step S13 as explanatory variables and the learning diagnosis result data acquired in step S13 as objective variables (step S14). In step S13, the learning data acquisition unit 23 acquires learning input data that does not include excluded data, so in step S14, the model generation unit 24 generates a learning model using the learning input data that does not include excluded data as explanatory variables. In step S14, the model generation unit 24 generates a learning model for each type of soil.

[0075] Examples of soil types include andosol, red soil, and immature sand dune soil. The effects of applying nitrogenous and phosphate fertilizers to these soils vary depending on the soil type. For example, the effect of applying nitrogenous fertilizer to red soil is smaller than the effect of applying the same amount of nitrogenous fertilizer to andosol. Furthermore, the effect of applying phosphate fertilizer to andosol is smaller than the effect of applying the same amount of phosphate fertilizer to red soil. Furthermore, immature sand dune soil has the property that when nitrogenous fertilizer is applied, it is easily washed away due to heavy rain, and excessive dryness can easily cause nitrogenous fertilizer concentration problems.

[0076] Therefore, for example, when learning input data for a field with black soil as an explanatory variable, the model generation unit 24 may weight the amount of phosphate fertilizer applied during learning. For example, when learning input data for a field with red soil as an explanatory variable, the model generation unit 24 may weight the amount of nitrogen fertilizer applied during learning. Furthermore, when learning input data for a field with immature dune soil as an explanatory variable, the model generation unit 24 may weight types of nitrogen fertilizer that are less likely to be washed away by rain during learning.

[0077] The learning device 2, for example, transmits the learning model generated in step S14 to the soil diagnostic device 1. The learning device 2, for example, transmits the generated learning model to the soil diagnostic device 1 via the network N. After the above processing, the learning device 2 ends the series of operations.

[0078] Next, the processing of the determination unit 22 in step S12 will be specifically described with reference to FIGS. 9(a) to 9(c). As described above, the determination unit 22 determines the diagnostic result data for learning to be used as a dependent variable in learning the learning model. The diagnostic result data for learning is generated by a plurality of diagnosticians based on other learning input data similar to the learning input data. The following description will mainly focus on a method in which the determination unit 22 identifies other learning input data similar to the learning input data and identifies the diagnostic result data for learning generated based on the other learning input data.

[0079] As shown in FIG. 9(a), in step S12, the determination unit 22 first acquires reference data for each of a plurality of reference fields and learning input data for a target field. The target field is a target field for which diagnosis result data for learning is determined. The determination unit 22 determines, for example, one field stored in the field database D1 as the target field. The determination unit 22 may determine one field as the target field, and the determination method is not particularly limited. The determination unit 22 then acquires learning input data for the determined target field.

[0080] 9(a), the determination unit 22 acquires the measurement results of the chemical composition of the soil and the measurement results of the three-phase distribution of the soil from the learning input data of the target field. The determination unit 22 acquires the measurement results of the chemical composition of the soil and the measurement results of the three-phase distribution of the soil from the reference data for each of the multiple reference fields. The determination unit 22 may acquire all of the numerical data that can be expressed numerically from the learning input data of the target field and the reference data for each of the multiple reference fields, or may acquire only a portion of the numerical data.

[0081] Next, the determination unit 22 identifies a reference field among the multiple reference fields that is closest to the state of the target field based on the acquired reference data and the learning input data of the target field. As shown in FIG. 9(b), the determination unit 22 performs dimensional compression on the acquired learning input data and reference data of the target field, thereby expressing each of the learning input data and reference data of the target field as coordinate points in a two-dimensional coordinate system. The determination unit 22 may perform dimensional compression using known methods such as principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), and uniform manifold approximation and projection (UMAP).

[0082] Next, as shown in FIG. 9(c), the determination unit 22 calculates the distance between a coordinate point of the learning input data of the target field and a coordinate point of the reference data (hereinafter, sometimes referred to as "the distance between two coordinate points"). The determination unit 22 calculates the distance between two coordinate points for each of a plurality of reference fields. For example, the x component of the coordinate point of the learning input data of the target field is set to x1, the x component of the coordinate point of the reference data is set to x1, the y component of the coordinate point of the learning input data of the target field is set to y1, the y component of the coordinate point of the reference data is set to y2, and the distance between the two coordinate points is set to T. The determination unit 22 may calculate the distance between the two coordinate points, for example, using the following formula (1): T=((x1-x2) 2 +(y1-y2) 2 ) 1 / 2 ···(1)

[0083] The coordinate points of the learning input data for a target field indicate, for example, the feature values ​​of the target field. The coordinate points of the reference data indicate, for example, the feature values ​​of the reference field. The closer two coordinate points are to each other, the closer the feature values ​​of the fields are to each other, and the closer the conditions of the fields are to each other. Conversely, the farther apart the two coordinate points are from each other, the more divergent the feature values ​​of the fields are to each other, and the more divergent the conditions of the fields are to each other.

[0084] For example, the determination unit 22 identifies the reference field in which the distance between the coordinate points of the learning input data of the target field and the coordinate points of the reference data is the shortest as the reference field in a state most similar to that of the target field. In the example of Figures 9(a) to 9(c), reference data for two reference fields (reference field A and reference field B) is shown. The determination unit 22 identifies reference field A as the reference field in a state most similar to that of the target field.

[0085] The determination unit 22 determines the learning input data of the identified reference field as another learning input data similar to the learning input data of the target field. The determination unit 22 determines the diagnostic result data of the identified reference field as the learning diagnostic result data of the target field. The determination unit 22 acquires the diagnostic result data of the identified reference field from the diagnostic result database D2. The determination unit 22 determines the acquired diagnostic result data of the reference field as the learning diagnostic result data of the target field.

[0086] In step S12, the determination unit 22 repeatedly executes the above-described process for all of the multiple fields stored in the field database D1. In other words, the determination unit 22 determines another field different from the one field stored in the field database D1 as the target field, and repeats the above-described process. After the above process, the determination unit 22 ends step S12.

[0087] Next, the effects of the soil diagnostic device 1, learning device 2, and soil diagnostic system 10 according to this embodiment will be described. In the soil diagnostic device 1 according to this embodiment, property data and farming data are input into a learning model to generate soil diagnostic result data. This learning model is trained using learning input data, including learning property data and learning farming data used by multiple assessors, as explanatory variables, and learning diagnostic result data generated by multiple assessors based on other learning input data similar to the learning input data as a target variable. Thus, by inputting property data and farming data into the learning model, diagnostic result data based on the knowledge of multiple assessors can be output. Therefore, soil diagnostic results can be obtained with high accuracy. The learning device 2 and soil diagnostic system 10 according to this embodiment achieve the same effects as the soil diagnostic device 1 described above.

[0088] The soil diagnostic device 1 further includes a notification unit 14 that notifies anomaly information indicating an abnormality in the soil based on the diagnostic result data generated by the generation unit 13. For example, if an abnormality is found in the soil compared to normal years, the soil abnormality can be notified to the diagnostician. This allows the cause of the soil abnormality to be quickly identified.

[0089] In the soil diagnostic device 1, the property data includes the measurement results of the soil's chemical composition and the measurement results of the soil's three-phase distribution. In this case, scientific measurement results are input to the learning model as property data, so diagnostic result data based on objective data can be generated.

[0090] In the soil diagnostic device 1, the farming data includes the results of agricultural practice of the previous crop in the field and the cultivation plan of the next crop in the field. In this case, for example, the state of the field from the viewpoint of the farmer of the field is input as the farming data, so that diagnostic result data can be generated based on not only scientific data but also the state of the field from the viewpoint of a person.

[0091] The learning device 2 further includes a determination unit 22 that determines the learning diagnostic result data for a target field based on reference data including property data and farming data for a plurality of preset reference fields and learning input data for a target field, which is the field for which the learning diagnostic result data is to be determined. For example, a good field, which is a field that has achieved good crop cultivation results, can be set as the reference field. In this case, the diagnostic result data for the good field can be determined as the learning diagnostic result data for the target field, and a learning model can be generated using the diagnostic result data for the good field as the objective variable. This makes it possible to generate a learning model that can generate diagnostic result data that brings the condition of the target field closer to that of a good field when property data and farming data are input. This allows for more accurate soil diagnostic results to be obtained.

[0092] In the learning device 2, the determination unit 22 identifies a reference field among multiple reference fields whose state is closest to the state of the target field based on the reference data and the learning input data for the target field, and determines the diagnostic result data of the identified reference field as the diagnostic result data for learning of the target field. For example, if a good field is set as the reference field, a good field whose state is closest to the state of the target field can be identified. Because the state of the identified good field is close to the state of the target field, it is thought that the amount of work required to bring the state of the target field closer to the good field state is small. Furthermore, a learning model can be generated using the diagnostic result data of the identified good field as the objective variable, so that when property data and farming data are input, a learning model can be generated that can generate diagnostic result data for bringing the state of the target field closer to the good field state with little work. Therefore, a diagnostic result for easily bringing the state of the field closer to the good field state can be obtained.

[0093] In the learning device 2, the determination unit 22 dimensionally compresses the learning input data and reference data of the target field, thereby expressing each of the learning input data and reference data of the target field as coordinate points in a two-dimensional coordinate system, and identifies the reference field whose state is closest to that of the target field based on the distance between the coordinate points of the learning input data and the coordinate points of the reference data of the target field. In this case, because the learning input data and reference data of the target field are dimensionally compressed, the amount of calculation required to determine the reference field whose state is closest to that of the target field can be reduced.

[0094] In the learning device 2, the learning data acquisition unit 23 acquires learning input data and learning diagnosis result data for each soil type, and the model generation unit 24 generates a learning model for each soil type. For example, fertilization designs for supplying nutrients necessary for crop cultivation may differ for each soil type. Since a learning model is generated for each soil type, it is possible to generate a learning model that can generate diagnosis result data according to the soil type.

[0095] In the learning device 2, the learning input data includes first data, which is learning input data used by the assessor to diagnose the soil at a certain point in time, and second data, which is learning input data used by the assessor to diagnose the soil at a point in time prior to the first point in time. The model generation unit 24 generates a learning model by weighting the first data more heavily than the second data for use in learning. For example, due to fluctuations in weather conditions, the first data used to diagnose the soil at a certain point in time is considered to better reflect the state of the field than the second data used to diagnose the soil at a point in time prior to the first point in time. By weighting the first data more heavily than the second data for use in learning, learning is performed with an emphasis on data closer to the point in time at which the learning model is generated. As a result, the learning model can be generated taking into account time-series changes in the state of the field, thereby obtaining more accurate soil diagnosis results.

[0096] The learning device 2 further includes a receiving unit 21 that receives the selection of excluded data, including learning property data or learning farming data that is not used in learning the learning model. The model generation unit 24 generates a learning model using the learning input data, excluding the excluded data, received by the receiving unit 21 as explanatory variables. For example, data including learning property data or farming data that is considered inappropriate for use in learning the learning model due to delays in farm work, farming errors, etc., can be selected as excluded data. In this case, the accuracy of learning can be improved by generating a learning model using learning input data, excluding the excluded data, as explanatory variables. Therefore, more accurate soil diagnostic results can be obtained.

[0097] Although the embodiment has been described in detail above, the present invention is not limited to the above embodiment. Modifications will be described below.

[0098] In the above embodiment, the determination unit 22 identified the reference field with the smallest distance between two coordinate points as the reference field whose state is closest to the state of the target field. However, the determination unit 22 may identify the reference field whose state is closest to the state of the target field based on the distance between the coordinate points of the learning input data of the target field and the coordinate points of the reference data. For example, the determination unit 22 may calculate a normalized value of the distance between the two coordinate points for each of multiple reference fields. The determination unit 22 may identify the reference field whose state is closest to the state of the target field based on the calculated value.

[0099] For example, the determination unit 22 acquires the maximum value among the x-component of the coordinate point of the learning input data of the target field, the x-component of the coordinate point of the learning input data of the target field, the x-component of the coordinate point of the reference data, and the y-component of the coordinate point of the reference data (hereinafter, "coordinate point components"). The determination unit 22 calculates a coefficient that, when multiplied by the acquired maximum value, is less than a predetermined value. The determination unit 22 may calculate a value obtained by dividing the predetermined value by the maximum value as the coefficient. The determination unit 22 multiplies each of the coordinate point components by the calculated coefficient to calculate normalized coordinate points. The determination unit 22 calculates the distance between two normalized coordinate points. For each of the multiple reference fields, the determination unit 22 calculates the product of the calculated distance, the x-component of the coordinate point of the reference data, and the y-component of the coordinate point of the reference data. The determination unit 22 identifies the reference field with the smallest calculated product as the reference field whose state is closest to that of the target field.

[0100] In the example of FIG. 9(b), all of the components of the coordinate point are positive values. However, if the components of the coordinate point include negative values, the determination unit 22 may correct each of the components of the coordinate point. For example, the determination unit 22 adds equal values ​​to each of the components of the coordinate point so that the smallest value among the components of the coordinate point is a predetermined positive value (e.g., 1). Based on each of the components of the coordinate point after addition, the determination unit 22 calculates the product of the calculated distance, the x-component of the coordinate point of the reference data, and the y-component of the coordinate point of the reference data for each of the multiple reference fields. The determination unit 22 identifies the reference field with the smallest calculated product as the reference field whose state is closest to that of the target field. In addition, if the components of the coordinate point include negative values, the determination unit 22 may calculate the product of the calculated distance, the x component of the coordinate point of the reference data, and the y component of the coordinate point of the reference data for each of the multiple reference fields, and identify the reference field with the smallest absolute value of the calculated product as the reference field whose state is closest to the state of the target field.

[0101] The determination unit 22 may identify a reference field whose state is closest to the state of the target field based on the geographical distance between the target field and the reference field. In this case, for example, the field database D1 stores learning input data including location information for each of a plurality of fields, and stores reference data including location information for each of a plurality of reference fields. The location information includes, for example, latitude and longitude. The determination unit 22 calculates the geographical distance between the target field and the reference field for each of the plurality of reference fields. The determination unit 22 identifies the reference field whose state is closest to the state of the target field based on the calculated geographical distance and the distance between two coordinate points. For example, for each of the plurality of reference fields, the determination unit 22 calculates the sum of the value obtained by multiplying the geographical distance by a predetermined value and the value obtained by multiplying the distance between two coordinate points by a predetermined value (i.e., the linear sum of the geographical distance and the distance between the two coordinate points). The determining unit 22 identifies the reference field with the smallest calculated sum as the reference field whose state is closest to that of the target field.

[0102] In the above embodiment, the model generation unit 24 identifies the generation order of the learning input data based on the time point information included in the learning input data, and weights the first data more heavily than the second data based on the identified generation order and uses the weighted first data for learning, thereby generating a learning model. However, the model generation unit 24 may weight the learning input data based on a criterion different from the generation order of the learning input data and use the weighted first data for learning. The model generation unit 24 may also perform learning without weighting the learning input data.

[0103] In the above embodiment, the determination unit 22 represents each of the learning input data and the reference data of the target field using coordinate points in a two-dimensional coordinate system by dimensionally compressing the learning input data and the reference data of the target field. However, the determination unit 22 may represent each of the learning input data and the reference data of the target field using coordinate points in a three-dimensional coordinate system. The dimensions of the coordinate system can be changed as appropriate.

[0104] In the above embodiment, the learning data acquisition unit 23 acquired learning input data and learning diagnosis result data for each type of soil. The model generation unit 24 generated a learning model for each type of soil. However, the learning data acquisition unit 23 may acquire learning input data and learning diagnosis result data for each type of crop cultivated in the field. The model generation unit 24 may generate a learning model for each type of crop.

[0105] Finally, various exemplary embodiments included in the present invention are described below in [E1] to [E12].

[0106] [E1] a property data acquisition unit that acquires property data relating to physical or chemical properties of the soil in the field; a farming data acquisition unit that acquires farming data related to farming in the field; a generation unit that inputs the property data and the farming data into a pre-generated learning model to generate soil diagnostic result data; The learning model is The learning input data including the learning property data and the learning farming data used in the soil diagnosis by a plurality of assessors are used as explanatory variables; A soil diagnostic device that is trained using learning diagnostic result data generated by the plurality of diagnosticians based on other learning input data similar to the learning input data as a response variable.

[0107] [E2] The soil diagnostic device according to [E1], further comprising a notification unit that notifies abnormality information indicating an abnormality in the soil based on the diagnostic result data generated by the generation unit.

[0108] [E3] The soil diagnostic device according to [E1] or [E2], wherein the property data includes at least one of a measurement result of the chemical composition of the soil and a measurement result of the three-phase distribution of the soil.

[0109] [E4] The soil diagnostic device according to any one of [E1] to [E3], wherein the farming data includes at least one of the agricultural practice results of a previous crop in the field and a cultivation plan for a subsequent crop in the field.

[0110] [E5] a learning data acquisition unit that acquires learning input data, including learning property data relating to the physical or chemical properties of the soil of a farm field and learning farming data relating to farming operations in the farm field, that has been used in soil diagnosis by a plurality of assessors, and learning diagnosis result data that has been generated by the plurality of assessors based on other learning input data similar to the learning input data; a model generation unit that generates a learning model by learning using the learning input data as an explanatory variable and the learning diagnostic result data as a target variable.

[0111] [E6] The learning device described in [E5] further includes a determination unit that determines the learning diagnostic result data of a target field based on reference data including the property data and the farming data of a plurality of preset reference fields and the learning input data of a target field, which is the field for which the learning diagnostic result data is to be determined.

[0112] [E7] The determination unit identifying, from among the plurality of reference fields, the reference field whose state is closest to the state of the target field based on the reference data and the learning input data of the target field; The learning device according to [E6], which determines the diagnosis result data of the identified reference field as the diagnosis result data for learning of the target field.

[0113] [E8] The determination unit expressing the learning input data and the reference data of the target field by coordinate points in a two-dimensional coordinate system by dimensionally compressing the learning input data and the reference data of the target field; A learning device described in [E7], which identifies the reference field whose state is closest to the state of the target field based on the distance between the coordinate points of the learning input data of the target field and the coordinate points of the reference data.

[0114] [E9] the learning data acquisition unit acquires the learning input data and the learning diagnosis result data for each type of soil, The learning device according to any one of [E5] to [E8], wherein the model generation unit generates the learning model for each type of soil.

[0115] [E10] The learning input data is First data is the learning input data used by the assessor to assess the soil at a certain point in time; and second data, which is the learning input data used by the assessor to assess the soil at a time point prior to the one time point, The learning device according to any one of [E5] to [E9], wherein the model generation unit generates the learning model by weighting the first data more heavily than the second data for use in learning.

[0116] [E11] a receiving unit that receives selection of excluded data including the learning property data or the learning farming data that is not used in learning the learning model, The learning device according to any one of [E5] to [E10], wherein the model generation unit generates the learning model using the learning input data that does not include the excluded data received by the reception unit as explanatory variables.

[0117] [E12] The soil diagnostic device according to any one of [E1] to [E4], A soil diagnostic system comprising the learning device according to any one of [E5] to [E11]. [Explanation of symbols]

[0118] 1...soil diagnostic device, 2...learning device, 10...soil diagnostic system, 11...property data acquisition unit, 12...farming data acquisition unit, 13...generation unit, 14...notification unit, 21...reception unit, 22...decision unit, 23...learning data acquisition unit, 24...model generation unit.

Claims

1. a property data acquisition unit that acquires property data relating to physical or chemical properties of the soil in the field; a farming data acquisition unit that acquires farming data related to farming in the field; a generation unit that inputs the property data and the farming data into a pre-generated learning model to generate soil diagnostic result data; The learning model is The learning input data including the learning property data and the learning farming data used in the soil diagnosis by a plurality of assessors are used as explanatory variables; A soil diagnostic device that is trained using learning diagnostic result data generated by the plurality of diagnosticians based on other learning input data similar to the learning input data as a response variable.

2. The soil diagnostic device according to claim 1 , further comprising a notification unit that notifies abnormality information indicating an abnormality in the soil based on the diagnostic result data generated by the generation unit.

3. The soil diagnostic device according to claim 1 or 2, wherein the property data includes at least one of a measurement result of a chemical composition of the soil and a measurement result of a three-phase distribution of the soil.

4. The soil diagnostic device according to claim 1 or 2, wherein the farming data includes at least one of an agricultural practice result of a previous crop in the field and a cultivation plan of a subsequent crop in the field.

5. a learning data acquisition unit that acquires learning input data, including learning property data relating to the physical or chemical properties of the soil of a farm field and learning farming data relating to farming operations in the farm field, that has been used in soil diagnosis by a plurality of assessors, and learning diagnosis result data that has been generated by the plurality of assessors based on other learning input data similar to the learning input data; a model generation unit that generates a learning model by learning using the learning input data as an explanatory variable and the learning diagnostic result data as a target variable.

6. The learning device described in claim 5, further comprising a determination unit that determines the learning diagnostic result data of a target field based on reference data including the property data and the farming data of a plurality of preset reference fields and the learning input data of a target field, which is the field for which the learning diagnostic result data is to be determined.

7. The determination unit identifying, from among the plurality of reference fields, the reference field whose state is closest to the state of the target field based on the reference data and the learning input data of the target field; The learning device according to claim 6 , wherein the diagnosis result data of the identified reference field is determined as the diagnosis result data for learning of the target field.

8. The determination unit expressing the learning input data and the reference data of the target field by coordinate points in a two-dimensional coordinate system by dimensionally compressing the learning input data and the reference data of the target field; The learning device according to claim 7 , wherein the reference field having a state closest to the state of the target field is identified based on the distance between the coordinate points of the learning input data of the target field and the coordinate points of the reference data.

9. the learning data acquisition unit acquires the learning input data and the learning diagnosis result data for each type of soil, The learning device according to claim 5 , wherein the model generation unit generates the learning model for each type of soil.

10. The learning input data is First data is the learning input data used by the assessor to assess the soil at a certain point in time; and second data, which is the learning input data used by the assessor to assess the soil at a time point prior to the one time point, The learning device according to claim 5 , wherein the model generation unit generates the learning model by weighting the first data more heavily than the second data for use in learning.

11. a receiving unit that receives selection of excluded data including the learning property data or the learning farming data that is not used in learning the learning model, The learning device according to claim 5 , wherein the model generation unit generates the learning model using the learning input data that does not include the excluded data and that is received by the reception unit as explanatory variables.

12. The soil diagnostic device according to claim 1 or 2; A soil diagnostic system comprising the learning device according to any one of claims 5 to 8.

Images