Soil component estimation device, soil component estimation method, and computer program

Abstract

To provide a technology to improve the estimation accuracy of soil components in a soil component estimation device.SOLUTION: A soil component estimation device includes an acquisition unit that acquires soil data including information on components of soil, a determination model generation unit that generates a determination model for determining whether or not the soil data acquired by the acquisition unit is normal, an estimation model generation unit that generates an estimation model for estimating the components of the soil from the soil data acquired by the acquisition unit, a determination unit that determines whether or not the soil data acquired by the acquisition unit is normal by using the determination model, and an estimation unit that uses the estimation model to estimate the components of the soil from the soil data determined as normal by the determination unit.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a soil component estimation device, a soil component estimation method, and a computer program.

Background Art

[0002] Conventionally, a soil component estimation device for estimating the components of soil has been known. For example, Patent Document 1 discloses a technique for generating an estimation model for estimating the components of soil from the soil type and water content ratio estimated from the measurement results of a soil sensor. Patent Document 2 discloses a technique for estimating the components of soil using the light reflection spectrum of soil.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, even with prior art such as Patent Documents 1 and 2, there is still room for improvement in the technique for improving the estimation accuracy of soil components in a soil component estimation device. In the technique of Patent Document 1, since the components of soil are estimated using the soil type, if the soil contains impurities such as straw, the estimation accuracy of the soil type may decrease, and the estimation accuracy of the soil components may decrease. In the technique of Patent Document 2, if foreign substances such as straw are contained, the light reflection spectrum of the soil may deviate greatly from the true value, and the estimation accuracy of the soil components may decrease.

[0005] The present invention has been made to solve the above-described problems, and an object thereof is to provide a technique for improving the estimation accuracy of soil components in a soil component estimation device. [Means for solving the problem]

[0006] The present invention has been made to solve at least some of the above-mentioned problems and can be realized in the following forms.

[0007] (1) According to one embodiment of the present invention, a soil component estimation device is provided. This soil component estimation device comprises: an acquisition unit that acquires soil data including information about the components of soil; a determination model generation unit that generates a determination model for determining whether the soil data acquired by the acquisition unit is normal or not; an estimation model generation unit that generates an estimation model for estimating the components of soil from the soil data acquired by the acquisition unit; a determination unit that uses the determination model to determine whether the soil data acquired by the acquisition unit is normal or not; and an estimation unit that uses the estimation model to estimate the components of soil from the soil data determined to be normal by the determination unit.

[0008] In this configuration, the determination unit uses a determination model to determine whether the soil data acquired by the acquisition unit is normal data. The estimation unit estimates the soil components for the soil data determined to be normal by the determination unit. For example, soil data that contains disturbances such as foreign matter, which would result in low estimation accuracy, can be excluded from the estimation target. Therefore, the accuracy of soil component estimation can be improved.

[0009] (2) In the soil component estimation device of the above form, a storage unit is provided that stores determination data including the correspondence between determined soil data, which has been determined to be normal or not, and the determination results of the determined soil data, and the determination model generation unit may generate the determination model by machine learning using the determination data. With this configuration, the storage unit stores determination data including the correspondence between determined soil data and its determination results. The determination model generation unit generates a determination model using, for example, soil data that has been determined to be normal from the determination data stored in the storage unit. As a result, even if the soil data acquired by the acquisition unit is unknown soil data, it is possible to determine whether it is normal or not by comparing it with determination data that reflects the results determined by, for example, an external analysis device, and thus make a highly accurate determination. Therefore, the determination accuracy of the determination model can be improved, and the estimation accuracy of soil components can be further improved.

[0010] (3) In the soil component estimation device of the above form, the judgment model generation unit may generate the judgment model by machine learning using a dataset obtained from the intermediate layer of a neural network for determining whether the soil data is normal or not. With this configuration, since the judgment model generation unit generates the judgment model by machine learning using a dataset obtained from the intermediate layer of a neural network, the dimension of the judgment model can be made smaller than the dimension of the soil data. This reduces the computational load and, for example, allows for the aggregation of the range in the data space that corresponds to a normal dataset to a smaller size, so that deviations from a normal dataset can be recognized with high accuracy. Therefore, the judgment accuracy of the judgment model can be further improved, and thus the accuracy of soil component estimation can be further improved.

[0011] (4) In the soil component estimation device of the above form, the determination model generation unit may generate a normal data region in the data space of the intermediate layer as the determination model using a normal dataset obtained from an intermediate layer connected to an output layer that outputs that the soil is normal, and the determination unit may determine whether the soil data is normal or not based on whether the dataset obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in the normal data region. With this configuration, the normal data region in the data space can be reduced in size and aggregated, thereby further improving the determination accuracy of the determination model.

[0012] (5) In the soil component estimation device of the above form, the determination model generation unit may generate the normal data region in the intermediate layer data space, which includes a set of datasets in the intermediate layer data space that are within a specific distance from the set of normal datasets, and the set of normal datasets. With this configuration, the determination unit determines that datasets in the intermediate layer data space that are relatively close in distance from the set of normal datasets are normal. That is, the determination unit uses a so-called outlier detection method to determine soil data that is not normal but is at a level where it would be acceptable to determine it as normal as normal, and the estimation unit estimates the components of the soil data that has been determined to be normal. This reduces the number of times soil data acquisition must be repeated, and thus soil components can be estimated relatively quickly.

[0013] (6) In the soil component estimation device of the above form, the determination model generation unit generates a plurality of abnormal data regions in the data space of the intermediate layer as the determination model using an abnormal data set obtained from an intermediate layer connected to an output layer that outputs that the soil is abnormal, and the determination unit determines the type of abnormality in the soil data by determining whether or not the data set obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in any of the plurality of abnormal data regions. With this configuration, the determination model generation unit generates a plurality of abnormal data regions using an abnormal data set obtained from an intermediate layer. The determination unit determines the type of abnormality in the soil data from the relationship between the data set obtained from the intermediate layer of the neural network into which the soil data to be determined is input and the plurality of abnormal data regions in the data space of the intermediate layer. As a result, the cause of the abnormality in the soil data that has been determined to be abnormal can be grasped, and by addressing the cause, the soil data can be acquired again and the soil components can be estimated.

[0014] (7) In the soil component estimation device of the above form, the acquisition unit acquires component data including the correspondence between measured soil data in which components have been measured and the measured components of the measured soil data, the estimation unit estimates the components of the measured soil from the measured soil data using the estimation model, and the estimation model generation unit may update the estimation model according to the magnitude of the difference between the estimated components of the measured soil data estimated by the estimation unit and the measured components included in the component data. With this configuration, the estimation unit estimates the components of the measured soil from measured soil data using the estimation model. For example, if the difference between the estimated components and the true values ​​of the components measured by an external analytical device is large, the estimation model generation unit updates the estimation model. This makes it possible to improve the estimation accuracy of the estimation model using the true values ​​of the components measured by an external analytical device, and thus further improve the estimation accuracy of soil components.

[0015] (8) According to another embodiment of the present invention, a soil component estimation method is provided in which a soil component estimation device estimates the components of soil. The soil component estimation method comprises: an acquisition step of acquiring soil data containing information about the components of soil; a determination model generation step of generating a determination model for determining whether the soil data is normal or not; an estimation model generation step of generating an estimation model for estimating the components of soil using the soil data; a determination step of determining whether the soil data acquired in the acquisition step is normal or not using the determination model; and an estimation step of estimating the components of soil from the soil data determined to be normal in the determination step using the estimation model. With this configuration, in the determination step, the determination model is used to determine whether the soil data acquired in the acquisition step is normal or not. In the estimation step, the components of soil are estimated from the soil data determined to be normal in the determination step. This makes it possible to exclude soil data from the estimation target that tends to have low estimation accuracy even if its components are estimated due to the inclusion of disturbances such as foreign matter, for example, and thus improve the estimation accuracy of soil components.

[0016] (9) According to yet another embodiment of the present invention, a computer program is provided that causes a computer to perform soil component estimation. The computer program causes the computer to perform the following functions: an acquisition function to acquire soil data containing information about the components of the soil; a determination model generation function to generate a determination model for determining whether the soil data is normal or not; an estimation model generation function to generate an estimation model for estimating the components of the soil using the soil data; a determination function to determine whether the soil data acquired by the acquisition function is normal or not using the determination model; and an estimation function to estimate the components of the soil from the soil data determined to be normal by the determination function using the estimation model. With this configuration, the computer uses the determination function to determine whether the soil data acquired by the acquisition function is normal or not using the determination model. For soil data determined to be normal by the determination function, the components of the soil are estimated by the estimation function. This makes it possible to exclude soil data from the estimation target that tends to have low estimation accuracy even if its components are estimated due to the inclusion of disturbances such as foreign matter, for example, and thus improve the estimation accuracy of soil components.

[0017] Furthermore, the present invention can be realized in various forms, for example, as a system including a soil component estimation device, a control method for such device and system, a computer program that causes such device and system to perform soil component estimation, a server device for distributing the computer program, a non-temporary storage medium storing the computer program, and so on. [Brief explanation of the drawing]

[0018] [Figure 1] This is a schematic diagram showing the general configuration of the soil component estimation system according to the first embodiment. [Figure 2] This is a flowchart illustrating the soil component estimation method of the first embodiment. [Figure 3] This is the first figure illustrating the soil component estimation method of the first embodiment. [Figure 4]It is the second figure for explaining the soil component estimation method of the first embodiment. [Figure 5] It is a flowchart for explaining the soil component estimation method of the second embodiment. [Figure 6] It is a figure for explaining the soil component estimation method of the second embodiment.

Mode for Carrying Out the Invention

[0019] <First Embodiment> FIG. 1 is a schematic diagram showing a schematic configuration of a soil component estimation system according to the first embodiment. The soil component estimation system 1 according to the present embodiment estimates, for example, the nutrient content (components) of soil collected from a field or the like. The system user of the soil component estimation system 1 adjusts the type and amount of fertilizer supplied to the field based on the nutrient content of the soil estimated by the soil component estimation system 1. Note that the field to which the soil component estimation system 1 is applied is not limited to the estimation of the nutrient content of soil in a field. For example, it may be applied to academic fields such as geological surveys for investigating soil characteristics.

[0020] The soil component estimation system 1 includes a measuring device 10 that measures the collected soil, a model generation device 20 that generates a model for estimating information regarding the soil collected by the measuring device 10, an analysis device 30 that estimates information regarding the soil using numerical values regarding the soil collected by the measuring device 10, and a display device 40 that displays the information regarding the soil estimated by the analysis device 30. In the present embodiment, these devices included in the soil component estimation system 1 are not aggregated in one place. For example, the measuring device 10 and the display device 40 are arranged at the site such as a field where the soil is collected, and the model generation device 20 and the analysis device 30 are arranged in a data center or the like for aggregating and analyzing data. The soil component estimation system 1 can estimate the nutrient content of the soil collected at the soil collection site by the measuring device 10, the display device 40, the model generation device 20, and the analysis device 30 exchanging information with each other through communication.

[0021] The measuring device 10 comprises a soil sensor 11 and a transmitting unit 12. The soil sensor 11 has a spectral spectrometer and the like. The soil sensor 11 measures spectral data in the visible light region and spectral data in the near-infrared region for the soil whose nutrient content is to be estimated (hereinafter referred to as "target soil for estimation"). The transmitting unit 12 is electrically connected to the soil sensor 11 and transmits the spectral data (soil data) measured by the soil sensor 11 to the model generation device 20.

[0022] The model generation device 20 has a computer comprising ROM, RAM, and a CPU, and includes a data input unit 21, a data holding unit 22, a judgment model generation unit 23, an estimation model generation unit 24, and a transmission / reception unit 25. In the model generation device 20, the CPU expands the computer program stored in ROM into RAM and executes it, thereby enabling the data holding unit 22, the judgment model generation unit 23, and the estimation model generation unit 24 to function.

[0023] The data input unit 21 is an input device for inputting information such as soil spectral data into the model generation device 20. In this embodiment, corresponding data is input, which includes spectral data of soil whose nutrient content has already been measured by an external analyzer (hereinafter referred to as "measured soil"), a determination result indicating whether or not the spectral data of the measured soil is normal, and the nutrient content of the measured soil.

[0024] The data storage unit 22 stores various data used to generate models in the judgment model generation unit 23 and estimation model generation unit 24, which will be described later. In this embodiment, the data stored includes the corresponding data of measured soil, real-time spectral data of the target soil transmitted from the measurement device 10, and the analysis results from the analysis device 30, all of which are input to the model generation device 20 via the data input unit 21.

[0025] The judgment model generation unit 23 generates a judgment model for determining whether the soil spectral data is normal or not, using the data stored in the data storage unit 22. In this embodiment, the judgment model generation unit 23 uses judgment data that combines the spectral data of the measured soil and the judgment result of whether the spectral data is normal or not, from the corresponding data of the measured soil stored in the data storage unit 22. The judgment model generation unit 23 generates a judgment model that indicates the normal range of spectral data by combining a neural network with outlier detection methods such as LOF, OC-SVM, and iForest. In this embodiment, the judgment model refers to a specific data region that exists in the data space. Details of the method for generating the judgment model will be described later.

[0026] The estimation model generation unit 24 generates an estimation model for estimating soil nutrient content from soil spectral data using data stored in the data storage unit 22. In this embodiment, the estimation model generation unit 24 uses component data from the corresponding data of measured soils stored in the data storage unit 22, which combines the spectral data of the measured soil with the true value data of the nutrient content of the soil having said spectral data. The estimation model generation unit 24 generates an estimation model using a neural network based on supervised machine learning using this component data. Note that the estimation model generation unit 24 is not limited to neural networks. It may also create an estimation model using regression methods such as Lasso regression, SVR (Support Vector Regression), XgBoost regression, or Random Forest regression.

[0027] The transmitting / receiving unit 25 is a communication device that exchanges information with the measuring device 10 and the analysis device 30. In this embodiment, the transmitting / receiving unit 25 receives spectral data of the target soil transmitted from the measuring device 10. The spectral data of the soil received by the transmitting / receiving unit 25 is transmitted to the analysis device 30 and stored in the data holding unit 22. The transmitting / receiving unit 25 transmits the judgment model generated by the judgment model generation unit 23 and the estimation model generated by the estimation model generation unit 24. The transmitting / receiving unit 25 receives various analysis results from the analysis device 30 transmitted from the analysis device 30. The analysis results from the analysis device 30 received by the transmitting / receiving unit 25 are stored in the data holding unit 22.

[0028] The analysis device 30 has a computer comprising ROM, RAM, and a CPU, and includes a transmitting / receiving unit 31, a determination unit 32, and an estimation unit 33. In the analysis device 30, the CPU expands the computer program stored in ROM into RAM and executes it, thereby functioning as the determination unit 32 and the estimation unit 33.

[0029] The transmitting / receiving unit 31 is a communication device that exchanges information with the model generation device 20 and the display device. In this embodiment, the transmitting / receiving unit 31 receives soil spectral data transmitted from the model generation device 20, and receives the judgment model generated by the judgment model generation unit 23 and the estimation model generated by the estimation model generation unit 24. The transmitting / receiving unit 31 also transmits the judgment results from the judgment unit 32 and the estimation results from the estimation unit 33 to the display device 40.

[0030] The determination unit 32 uses the determination model generated by the determination model generation unit 23 to determine whether the soil spectral data is normal or not. In this embodiment, the determination unit 32 determines whether the soil spectral data is normal or not by determining whether the soil spectral data is included in a specific data area in the data space generated by the determination model generation unit 23.

[0031] The estimation unit 33 uses the estimation model generated by the estimation model generation unit 24 to estimate soil nutrients, acidity, and other properties from the soil spectral data. In this embodiment, the estimation unit 33 estimates the soil nutrients by adding appropriate weights to each value of the spectral data of soil determined to be normal by the determination unit 32 and synthesizing them. In other words, the estimation model in this embodiment refers to the weighting values ​​calculated by the neural network.

[0032] The display device 40 comprises a receiving unit 41 and a display 42. The display device 40 displays the determination results from the determination unit 32 and the estimation results from the estimation unit 33, which are received by the receiving unit 41, on the display 42. This allows the system user to know the soil measurement results.

[0033] Next, the details of the soil component estimation method of this embodiment will be described. The soil component estimation method of this embodiment is started at any time at the discretion of the system user, for example, when information on soil nutrient content is needed, by operating a start button (not shown) on the soil component estimation system 1.

[0034] Figure 2 is a flowchart illustrating the soil component estimation method of this embodiment. In the soil component estimation method of this embodiment, first, spectral data of the target soil is measured (step S11). Specifically, the measuring device 10 collects soil from a field using a soil sampling device (not shown), and measures the spectral data of the collected soil using a spectral spectrometer or the like that of the soil sensor 11. The measured spectral data is sent to the model generation device 20 via the transmitting unit 12. The spectral data of the soil sent to the model generation device 20 is sent to the analysis device 30 via the transmitting / receiving unit 25 of the model generation device 20, and is also stored in the data holding unit 22 of the model generation device 20.

[0035] Next, it is determined whether the measured spectral data is normal or not (step S12). Specifically, the analysis device 30 uses a judgment model to determine whether the spectral data of the target soil transmitted from the measurement device 10 via the model generation device 20 is normal or not.

[0036] Figure 3 is the first diagram illustrating the determination method included in the soil component estimation method of this embodiment. Figure 3 shows a schematic diagram of a neural network for preprocessing the spectral data of the target soil when the analysis device 30 generates a determination model. The neural network NN1 shown in Figure 3 has an input layer into which the spectral data of the target soil is input, and an output layer that outputs the determination result, with an intermediate layer that performs various calculations on the spectral data of the target soil. In this embodiment, in step S12, the dataset obtained from the intermediate layer shown in Figure 3 is used to determine whether the spectral data of the soil acquired in step S11 is normal or not.

[0037] Figure 4 is a second diagram illustrating the determination method included in the soil component estimation method of this embodiment. Figure 4 schematically shows the data space of the intermediate layer of the neural network NN1. The dataset obtained from the intermediate layer of the neural network NN1 includes a normal dataset obtained from an intermediate layer connected to an output layer that outputs that the soil spectral data is normal, and an abnormal dataset obtained from an intermediate layer connected to an output layer that outputs that the soil spectral data is abnormal. In the data space Ds1 of Figure 4, a set of normal datasets G1 and a set of abnormal datasets G2 are shown.

[0038] In step S12, the dataset obtained from the hidden layer of the neural network when the soil spectral data acquired in step S11 is input to the neural network (hereinafter referred to as the "hidden layer dataset") is plotted in the data space Ds1 shown in Figure 4. For example, suppose that the hidden layer dataset Md1 is plotted in the data space Ds1 as shown in Figure 4. In this case, since the hidden layer dataset Md1 is included in the set of normal datasets G1, the determination unit 32 determines that the soil spectral data corresponding to the hidden layer dataset Md1 is normal. Also, suppose that the hidden layer dataset Md2 is plotted on the data space Ds1. In this case, since the hidden layer dataset Md2 is included in the set of abnormal datasets G2, the determination unit 32 determines that the soil spectral data corresponding to the hidden layer dataset Md2 is abnormal.

[0039] Furthermore, in this embodiment, the determination unit 32 determines whether the soil spectral data is normal or not based on the distance between the set of normal datasets G1 and the intermediate layer dataset in the data space shown in Figure 4. Specifically, the data space Ds1 includes the set of normal datasets G1, and the range from the outer edge of the set of normal datasets G1 to a specific distance (threshold) is set as the normal data area AG1 (see Figure 4). For example, as shown in Figure 4, if the intermediate layer dataset Md3 is plotted inside the normal data area AG1, that is, if the distance between the set of normal datasets G1 and the intermediate layer dataset Md3 is relatively small, the determination unit 32 determines that the soil spectral data corresponding to the intermediate layer dataset Md3 is normal. On the other hand, if the intermediate layer dataset Md4 is plotted outside the normal data area AG1, that is, if the distance between the set of normal datasets G1 and the intermediate layer dataset Md4 is relatively large, the determination unit 32 determines that the soil spectral data corresponding to the intermediate layer dataset Md4 is abnormal.

[0040] In this embodiment, regions in the data space, such as the normal data region AG1 shown in Figure 4, are determined by outlier detection methods such as LOF, OC-SVM, and iForest. In step S12, a method combining a neural network and an outlier detection method with a set threshold for the distance to the dataset is used to determine whether the soil spectral data is normal or not.

[0041] In step S12, if the determination unit 32 determines that the soil spectral data is normal (step S12: Yes), the analysis device 30 estimates the soil nutrient content from the soil spectral data (step S13). Specifically, the estimation unit 33 estimates the soil nutrient content by applying appropriate weights to each value of the spectral data of the soil determined to be normal by the determination unit 32 and synthesizing them. The numerical value indicating the soil nutrient content estimated by the estimation unit 33 is transmitted to the receiving unit 41 of the display device 40 via the transmitting / receiving unit 31 of the analysis device 30. The display device 40 displays the numerical value indicating the soil nutrient content transmitted to the receiving unit 41 on the display 42 (step S14). This allows the system user to know the estimated value of the soil nutrient content collected by the measuring device 10.

[0042] In step S12, if the determination unit 32 determines that the soil spectral data is abnormal (step S12: No), the display device 40 displays the determination result (step S15). Specifically, the determination result of the determination unit 32 is transmitted to the receiving unit 41 of the display device 40 via the transmitting / receiving unit 31. The display device 40 displays the determination result received by the receiving unit 41 on the display 42. This allows the system user to recognize that the measurement by the measuring device 10 was unsuccessful. In this embodiment, in step S15, the display 42 outputs a message to the system user prompting them to remeasure the soil spectral data, along with the determination result that the soil spectral data is abnormal. This allows the system user to improve the expected cause of the abnormality and perform a remeasurement.

[0043] Next, several methods for updating the judgment model or estimation model in the soil component estimation system 1 of this embodiment will be described. In the soil component estimation system 1 of this embodiment, the data holding unit 22 stores corresponding data of measured soil and real-time soil spectral data transmitted from the measuring device 10. In the soil component estimation system 1, when the data in the data holding unit 22 is updated, the judgment model or estimation model is updated. This further improves the accuracy of nutrient estimation by the soil component estimation system 1.

[0044] In the first update method, when component data combining measured soil spectral data and true nutrient data is newly input to the model generation device 20, the estimation unit 33 estimates the nutrient content using the measured soil spectral data. The estimation model generation unit 24 updates the estimation model according to the magnitude of the difference between the estimated nutrient content estimated by the estimation unit 33 and the true nutrient data. Specifically, the estimation model generation unit 24 updates the estimation model when the difference value is greater than a certain threshold.

[0045] In the second update method, when component data combining the spectral data of the measured soil and the true value data of nutrients is newly input to the model generation device 20, the determination unit 32 makes a determination on the spectral data of the measured soil. If the determination unit 32 determines that the spectral data of the measured soil is normal, the estimation model generation unit 24 updates the estimation model using the component data of the spectral data of the measured soil that was the subject of the determination unit 32's determination.

[0046] In the third update method, when component data combining the spectral data of the measured soil and the true value data of nutrients is newly input to the model generation device 20, the determination unit 32 makes a determination on the spectral data of the measured soil. The display device 40 notifies the system user of the determination result by displaying the determination result from the determination unit 32 on the display 42. The system user, knowing the determination result from the determination unit 32, sets a flag manually, and the determination model generation unit 23 updates the determination model according to the flag setting.

[0047] In the fourth update method, when real-time spectral data of the target soil is transmitted from the measuring device 10, the determination unit 32 makes a determination on the real-time spectral data of the target soil (step S12 in Figure 2). The display device 40 notifies the system user of the determination result (normal or abnormal) from the determination unit 32 by displaying it on the display 42 (step S14 or step S15 in Figure 2). The system user, knowing the determination result from the determination unit 32, manually sets a flag, and the determination model generation unit 23 updates the determination model according to the flag setting.

[0048] In the fifth update method, when real-time spectral data of the target soil is transmitted from the measuring device 10, the estimation unit 33 estimates the soil's nutritional value, for example, as in step S13 of the soil component estimation method described above. The determination unit 32 determines whether the soil's nutritional value estimated by the estimation unit 33 falls within the data range of soil nutritional value stored in the data holding unit 22. The determination unit 32 sets a normal flag if the estimated soil's nutritional value falls within the data range of the data holding unit 22, and sets an abnormal flag if the estimated soil's nutritional value falls outside the data range of the data holding unit 22. The determination model generation unit 23 updates the determination model according to the set flag.

[0049] As described above, according to the soil component estimation system 1 of this embodiment, the determination unit 32 uses a determination model to determine whether the soil spectral data acquired by the measuring device 10 is normal data. The estimation unit 33 estimates the soil's nutrient content for the soil spectral data that the determination unit 32 has determined to be normal. For example, it can exclude from the estimation target soil spectral data that would result in low estimation accuracy due to the inclusion of disturbances such as foreign matter. Therefore, the estimation accuracy of soil components can be improved.

[0050] Furthermore, according to the soil component estimation system 1 of this embodiment, the data holding unit 22 stores judgment data that includes the correspondence between the spectral data of the measured soil and the judgment result. The judgment model generation unit 23 generates a judgment model using the spectral data of the measured soil that has been determined to be normal from the judgment data stored in the data holding unit 22. As a result, even if the spectral data of the soil acquired by the measuring device 10 is spectral data of an unknown soil, it is possible to determine whether or not it is normal by comparing it with judgment data that reflects the results determined by, for example, an external analytical device, thereby enabling highly accurate judgment. Therefore, the judgment accuracy of the judgment model can be improved, and the accuracy of nutrient estimation can be further improved.

[0051] Furthermore, according to the soil component estimation system 1 of this embodiment, the judgment model generation unit 23 generates a judgment model by machine learning using a dataset obtained from the intermediate layer of the neural network NN1. This makes the dimension of the judgment model smaller than the dimension of the soil spectral data, thereby reducing the computational load and allowing the set of normal datasets G1 in the data space Ds1 shown in Figure 4 to be aggregated into a smaller size. Therefore, deviations from the set of normal datasets G1 can be recognized with high accuracy, further improving the accuracy of soil component estimation.

[0052] Furthermore, according to the soil component estimation system 1 of this embodiment, the determination unit 32 determines that the spectral data of the soil is normal for datasets that are relatively close to the set of normal datasets G1 in the intermediate layer data space. In other words, the determination unit 32 uses a so-called outlier detection method to use spectral data of soil that is not normal but is at a level that can be judged as normal for estimating nutrient levels. This reduces the number of times soil spectral data acquisition has to be repeated, and thus enables relatively rapid estimation of soil nutrient levels.

[0053] Furthermore, according to the soil component estimation system 1 of this embodiment, the estimation unit 33 uses an estimation model to estimate the nutrient content of the soil from the spectral data of the soil whose nutrient content has been measured. For example, if the difference between this estimated nutrient content and the true value of the nutrient content measured by an external analytical device is large, the estimation model generation unit 24 updates the estimation model. This improves the estimation accuracy of the estimation model by using the true value of the nutrient content measured by an external analytical device, thereby further improving the accuracy of nutrient content estimation.

[0054] Furthermore, according to the soil component estimation method of this embodiment, in step S12, the soil component estimation system 1 uses a judgment model to determine whether the soil spectral data acquired in step S11 is normal data. In step S13, the soil component estimation system 1 estimates the soil nutrient content for the soil spectral data that was determined to be normal in step S12. This makes it possible to exclude from the estimation target soil spectral data that would otherwise have low estimation accuracy due to the inclusion of disturbances such as foreign matter, thereby improving the accuracy of nutrient content estimation.

[0055] Furthermore, according to the computer program that causes a computer to perform soil component estimation in this embodiment, the soil component estimation system 1 uses the judgment function of the analysis device 30 to determine whether the soil spectral data acquired by the acquisition function of the measurement device 10 is normal data, using a judgment model. For soil spectral data that the judgment function of the analysis device 30 determines to be normal, the estimation function of the analysis device 30 estimates the soil's nutrient content. This makes it possible to exclude from the estimation target soil spectral data that tends to have low estimation accuracy due to the inclusion of disturbances such as foreign matter, thereby improving the accuracy of nutrient content estimation.

[0056] <Second Embodiment> Figure 5 is a flowchart illustrating the soil component estimation method of the second embodiment. The soil component estimation device of the second embodiment differs from the soil component estimation device of the first embodiment (Figure 1) in that the determination unit determines the type of abnormality.

[0057] In the soil component estimation method of the second embodiment, first, similar to the soil component estimation method of the first embodiment, spectral data of the target soil is measured (step S11), and it is determined whether the measured spectral data is normal or not (step S12). In step S12, if the determination unit 32 determines that the soil spectral data is normal (step S12: Yes), the analysis device 30 estimates the nutrient content from the soil spectral data (step S13). On the other hand, in step S12, if the determination unit 32 determines that the soil spectral data is abnormal (step S12: No), the process proceeds to step S25.

[0058] In step S25, the determination unit 32 determines the type of abnormality using the soil spectral data (step S25). In this embodiment, the determination unit 32 of the analysis device 30 uses the soil spectral data to determine the cause of the abnormality in the spectral data.

[0059] Figure 6 illustrates the determination method provided by the soil component estimation method of this embodiment. Figure 6 shows the data space of the intermediate layer of the neural network. The set of datasets obtained from the intermediate layer of the neural network includes a set of normal datasets G1, as well as sets of multiple abnormal datasets G21, G22, and G23.

[0060] When estimating soil nutrient content using soil spectral data, "types of anomalies" that negatively affect component estimation include ambient light, soil clogging, and straw clogging. In the data space of Figure 6, each of the anomaly dataset sets G21, G22, and G23 corresponds to each of these anomalies. For example, the anomaly dataset set G21 includes datasets obtained from the intermediate layer corresponding to the spectral data of soil determined to be anomaly due to ambient light. The anomaly dataset set G22 includes datasets obtained from the intermediate layer corresponding to the spectral data of soil determined to be anomaly due to soil clogging. The anomaly dataset set G23 includes datasets obtained from the intermediate layer corresponding to the spectral data of soil determined to be anomaly due to straw clogging. Furthermore, in this embodiment, anomaly data regions AG21, AG22, and AG23 are set, each containing each of the anomaly dataset sets G21, G22, and G23, and extending to a specific distance from the outer edges of each of the anomaly dataset sets G21, G22, and G23 (see Figure 6).

[0061] In step S25, the intermediate layer data is plotted on the data space shown in Figure 6. For example, suppose the intermediate layer dataset Md5 is plotted on the data space Ds2. In this case, as shown in Figure 6, the intermediate layer dataset Md5 is located inside the anomalous data region AG21 which includes the set of anomalous datasets G21. As a result, the determination unit 32 determines that the soil spectral data corresponding to the intermediate layer dataset Md5 is anomalous, and that the anomalousness is due to ambient light. The determination unit 32 similarly determines the anomalous sets G22 and G23, and the anomalous data regions AG22 and AG23, respectively, according to their positional relationship with the intermediate layer data. In this way, step S25 determines the type of anomalousness for the intermediate layer data that was determined to be anomalous in step S12. Furthermore, if the intermediate layer dataset plotted on the data space Ds2 is not located inside any of the sets of anomalous datasets G21, G22, G23, or the anomalous data regions AG21, AG22, AG23, the determination unit 32 determines that the type of anomalous condition is one other than the above-mentioned ambient light, soil clogging, and straw clogging.

[0062] Next, the display device 40 displays the type of abnormality (step S26). Specifically, the determination result regarding the type of abnormality by the determination unit 32 is transmitted to the receiving unit 41 of the display device 40 via the transmitting / receiving unit 31. The determination result received by the receiving unit 41 of the display device 40 is displayed on the display 42. This allows the system user to recognize that the measurement by the measuring device 10 has failed and what kind of failure it was.

[0063] As described above, according to the soil component estimation system of this embodiment, the judgment model generation unit 23 generates multiple anomaly data regions AG21, AG22, and AG23 corresponding to each of the multiple types of anomalies using an anomaly dataset obtained from the intermediate layer. The judgment unit 32 determines the type of anomaly in the soil spectral data from the relationship between the dataset obtained from the intermediate layer of the neural network, which is input with spectral data of the soil to be judged in the intermediate layer data space Ds2, and the multiple anomaly data regions AG21, AG22, and AG23. This makes it possible to understand the cause of the anomaly in the spectral data of soil that has been determined to be anomaly, and by addressing that cause, the spectral data of the soil can be acquired again and the nutrient content of the soil can be estimated.

[0064] <Modified form of this embodiment> The present invention is not limited to the embodiments described above, and can be implemented in various forms without departing from its spirit, for example, the following modifications are also possible.

[0065] [Example 1] In the above embodiment, the soil component estimation system 1 uses soil spectral data to determine whether the soil is normal or not, and to estimate its nutrient content. However, the soil data is not limited to this. Any physical quantity related to soil characteristics, such as soil moisture content, permeability, electrical conductivity, pH, and particle size distribution, may be used. Furthermore, the soil component estimation system 1 does not only estimate soil nutrients. It may also estimate other soil components or soil characteristics as described above.

[0066] [Differentiation 2] In the embodiment described above, the input data used was the spectral data of the soil acquired by the soil sensor 11. However, the determination may be made using data obtained by processing the acquired spectral data, rather than using the spectral data acquired by the soil sensor 11 directly. For example, data normalized to a mean of 0 and a variance of 1 may be used, or data to which a Savitzky-Golay filter has been applied may be used.

[0067] [Difference 3] In the above embodiment, a method combining a neural network and an outlier detection method in which a threshold distance to the dataset is set is used to determine whether the soil spectral data is normal or not. However, the method for determining whether the soil spectral data is normal or not is not limited to this. For example, it is not necessary to combine the outlier detection method with data preprocessing using a neural network.

[0068] [Differentiation Example 4] In the above-described embodiment, the nutrient content of the soil was estimated by adding appropriate weights to each value of the soil's spectral data and synthesizing them. However, the method for determining whether the soil's spectral data is normal or the method for estimating the nutrient content are not limited to this.

[0069] [Difference 5] In the above-described embodiment, the measuring device 10, model generation device 20, analysis device 30, and display device 40 of the soil component estimation system 1 were not integrated into one location. However, the measuring unit, model generation unit, analysis unit, and display unit may be provided as a single device.

[0070] The embodiments of this specification have been described above based on the embodiments and modifications described above. The embodiments described above are for the purpose of facilitating understanding of this specification and do not limit it. This specification may be modified and improved without departing from its spirit and the scope of the claims, and equivalents thereof are included in this specification. Furthermore, any technical features that are not described as essential in this specification may be deleted as appropriate.

[0071] [Application Example 1] A soil composition estimation device, An acquisition unit that acquires soil data including information about soil composition, A determination model generation unit generates a determination model for determining whether the soil data acquired by the acquisition unit is normal or not, The acquisition unit generates an estimation model for estimating soil components from the soil data acquired by the acquisition unit, A determination unit that determines whether the soil data acquired by the acquisition unit is normal or not using the aforementioned determination model, The system includes an estimation unit that uses the estimation model to estimate the soil components from soil data determined to be normal by the determination unit, and Soil composition estimation device. [Application Example 2] The soil component estimation device described in Application Example 1 further, The system includes a storage unit that stores determination data, which includes the correspondence between determined soil data (whether it is normal or not) and the determination results of the determined soil data. The judgment model generation unit generates the judgment model by machine learning using the judgment data. Soil composition estimation device. [Application Example 3] A soil component estimation device as described in Application Example 1 or Application Example 2, The judgment model generation unit generates the judgment model by machine learning using a dataset obtained from the intermediate layer of a neural network for determining whether the soil data is normal or not. Soil composition estimation device. [Application Example 4] A soil component estimation device described in any one of Application Examples 1 to 3, The judgment model generation unit generates the judgment model using the normal data set obtained from the intermediate layer connected to the output layer which outputs that the soil is normal, and the normal data region in the data space of the intermediate layer. The determination unit determines whether the soil data is normal or not based on whether the dataset obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in the normal data area. Soil composition estimation device. [Application Example 5] A soil component estimation device described in any one of Application Examples 1 to 4, The determination model generation unit generates the normal data region in the intermediate layer data space, which includes a set of datasets whose distance from the set of normal datasets is within a specific distance, and the set of normal datasets. Soil composition estimation device. [Application Example 6] A soil component estimation device described in any one of Application Examples 1 to 5, The judgment model generation unit generates a judgment model by using an anomaly dataset obtained from an intermediate layer connected to an output layer that outputs that the soil is abnormal, and by using multiple anomaly data regions in the data space of the intermediate layer. The determination unit determines the type of anomaly in the soil data based on whether the dataset obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in any of the multiple anomaly data regions. Soil composition estimation device. [Application Example 7] A soil component estimation device described in any one of Application Examples 1 to 6, The acquisition unit acquires component data that includes the correspondence between measured soil data in which components have been measured and the measured components of the measured soil data. The estimation unit uses the estimation model to estimate the measured soil components from the measured soil data. The estimation model generation unit updates the estimation model according to the magnitude of the difference between the estimated components of the measured soil data estimated by the estimation unit and the measured components included in the component data. Soil composition estimation device. [Application Example 8] A soil component estimation method in which a soil component estimation device estimates the components of soil, The acquisition process involves obtaining soil data, including information about the composition of the soil. A determination model generation step for generating a determination model for determining whether the soil data is normal or not, An estimation model generation step, which generates an estimation model for estimating the components of the soil using the soil data, A determination step is to determine whether the soil data acquired in the acquisition step is normal or not using the determination model described above, The system includes an estimation step of using the estimation model to estimate the soil components from soil data that was determined to be normal in the determination step, Methods for estimating soil composition. [Application Example 9] A computer program that causes a computer to perform soil composition estimation, A function to acquire soil data, including information about soil composition, A determination model generation function generates a determination model for determining whether the soil data is normal or not, An estimation model generation function that generates an estimation model for estimating the components of the soil using the aforementioned soil data, A determination function that uses the aforementioned determination model to determine whether the soil data acquired by the acquisition function is normal or not, Using the estimation model, the computer is instructed to perform an estimation function that estimates the soil components from soil data that has been determined to be normal by the judgment function. Computer program. [Explanation of Symbols]

[0072] 1…Soil composition estimation system 11…Soil Sensor 21...Data Entry Section 22...Data storage unit 23... Judgment Model Generation Unit 24…Estimation Model Generation Unit 32…Judgment section 33…Estimation part AG1...Normal data area AG21, AG22, AG23... Abnormal data area Ds1, Ds2…Data space NN1...Neural Network G1…Set of normal datasets G21, G22, G23... A collection of anomalous datasets

Claims

1. A soil composition estimation device, An acquisition unit that acquires soil data including information about soil composition, A determination model generation unit generates a determination model for determining whether the soil data acquired by the acquisition unit is normal or not, The acquisition unit generates an estimation model for estimating soil components from the soil data acquired by the acquisition unit, A determination unit that determines whether the soil data acquired by the acquisition unit is normal or not using the aforementioned determination model, The system includes an estimation unit that uses the estimation model to estimate the soil components from soil data determined to be normal by the determination unit, and Soil composition estimation device.

2. The soil component estimation device according to claim 1 further, The system includes a storage unit that stores determination data, which includes the correspondence between determined soil data (whether it is normal or not) and the determination results of the determined soil data. The judgment model generation unit generates the judgment model by machine learning using the judgment data. Soil composition estimation device.

3. A soil component estimation device according to claim 1 or claim 2, The judgment model generation unit generates the judgment model by machine learning using a dataset obtained from the intermediate layer of a neural network for determining whether the soil data is normal or not. Soil composition estimation device.

4. A soil component estimation device according to claim 3, The judgment model generation unit generates the judgment model using the normal data set obtained from the intermediate layer connected to the output layer which outputs that the soil is normal, and the normal data region in the data space of the intermediate layer. The determination unit determines whether the soil data is normal or not based on whether the dataset obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in the normal data area. Soil composition estimation device.

5. A soil component estimation device according to claim 4, The determination model generation unit generates the normal data region in the intermediate layer data space, which includes a set of datasets whose distance from the set of normal datasets is within a specific distance, and the set of normal datasets. Soil composition estimation device.

6. A soil component estimation device according to claim 3, The judgment model generation unit uses an anomaly dataset obtained from an intermediate layer connected to an output layer that outputs that the soil is abnormal, to generate a plurality of anomaly data regions in the data space of the intermediate layer as the judgment model. The determination unit determines the type of anomaly in the soil data based on whether the dataset obtained from the intermediate layer of the neural network into which the soil data to be determined is input is included in any of the multiple anomaly data regions. Soil composition estimation device.

7. A soil component estimation device according to claim 1 or claim 2, The acquisition unit acquires component data that includes the correspondence between measured soil data in which components have been measured and the measured components of the measured soil data. The estimation unit uses the estimation model to estimate the measured soil components from the measured soil data. The estimation model generation unit updates the estimation model according to the magnitude of the difference between the estimated components of the measured soil data estimated by the estimation unit and the measured components included in the component data. Soil composition estimation device.

8. A soil component estimation method in which a soil component estimation device estimates the components of soil, The acquisition process involves obtaining soil data, including information about the composition of the soil. A determination model generation step for generating a determination model for determining whether the soil data is normal or not, An estimation model generation step, which generates an estimation model for estimating the components of the soil using the soil data, A determination step is to determine whether the soil data acquired in the acquisition step is normal or not using the determination model described above, The system includes an estimation step of using the estimation model to estimate the soil components from soil data that was determined to be normal in the determination step, Methods for estimating soil composition.

9. A computer program that causes a computer to perform soil composition estimation, A function to acquire soil data, including information about soil composition, A determination model generation function generates a determination model for determining whether the soil data is normal or not, An estimation model generation function that generates an estimation model for estimating the components of the soil using the aforementioned soil data, A determination function that uses the aforementioned determination model to determine whether the soil data acquired by the acquisition function is normal or not, Using the estimation model, the computer is instructed to perform an estimation function that estimates the soil components from soil data that has been determined to be normal by the judgment function. Computer program.

Images