Lodging estimation system and lodging estimation method

Abstract

To enable easy estimation of a lodging condition of plants cultivated in a field, even when the field is large in scale, without increasing human labor costs.SOLUTION: A lodging estimation system comprises: an acquired image data acquisition unit that acquires acquired image data obtained by imaging a field; an on-site survey data acquisition unit that acquires lodging index values that are results of an on-site survey of a lodging state of plants grown in the field at a time corresponding to a time when the acquired image data is measured; and a lodging estimation unit that estimates, based on estimated-target acquired image data obtained by measuring an estimated-target field, a lodging index value on the basis of a relationship between the acquired image data and the lodging index values obtained as actual results through the acquired image data and the on-site survey.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a lodging estimation system and a lodging estimation method. [Background technology]

[0002] Plants grown in a field may grow excessively due to factors such as excessive fertilization. Excessive growth can cause the crop stems to bend and fall over at harvest time. This phenomenon is called lodging, and is known to cause a decline in crop quality, a decrease in harvest yield, and combine damage during harvesting. Patent Document 1 discloses a technology that calculates an index indicating the degree of lodging of agricultural crops from the combine speed, utilizing the fact that the greater the degree of lodging, the slower the combine speed. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-170999 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-mentioned technology is inconvenient because it is not possible to calculate the degree of lodging without using a combine harvester. On the other hand, visually checking the lodging state of plants by hand is not practical in large-scale fields because of the increased labor costs.

[0005] The present invention has been made in consideration of these circumstances, and its purpose is to provide a lodging estimation system and lodging estimation method that can easily estimate the lodging state of plants growing in a field, even in a large-scale field, without increasing human costs. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, one aspect of the present invention is a lodging estimation system having an acquired image data acquisition unit that acquires acquired image data obtained by photographing a field, a field survey data acquisition unit that acquires a lodging index value that is the result of a field survey of the lodging state of plants grown in the field at a time corresponding to the time when the acquired image data was measured, and a lodging estimation unit that estimates a lodging index value based on acquired image data of an estimation target measured in the field of the estimation target, based on the relationship between the acquired image data and the lodging index value obtained as an actual result from the field survey.

[0007] Another aspect of the present invention is a lodging estimation method performed by a computer used in a lodging estimation system, in which an image data acquisition unit acquires image data obtained by photographing a field, and acquires a lodging index value that is the result of a field survey of the lodging state of plants grown in the field at a time corresponding to the time when the acquired image data was measured, and a lodging estimation unit estimates a lodging index value based on the acquired image data of the estimation target obtained by measuring the field of the estimation target, based on the relationship between the acquired image data and the lodging index value obtained as an actual result from the field survey. [Effects of the Invention]

[0008] As described above, according to the present invention, it is possible to easily estimate the lodging state of plants growing in a farm field, even in a large-scale farm field, without increasing personnel costs. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a schematic block diagram showing the configuration of a falling estimation system 1 according to an embodiment of the present invention. [Figure 2] 10 is a diagram showing an example of field survey data stored in a field survey data storage unit 102. FIG. [Figure 3] FIG. 3 is a diagram showing an example of vegetation index information stored in a vegetation index information storage unit 105. [Figure 4] FIG. 10 is a diagram showing an example of a vegetation index image based on a vegetation index calculated from acquired image data. [Figure 5] FIG. 2 is a diagram showing an example of model information stored in a model information storage unit 108. [Figure 6] 10 is a diagram showing an example of a lodging map generated by a lodging map generating unit 121. FIG. [Figure 7] 10 is a flowchart illustrating a process flow in which the lodging estimation system 1 generates a learning model. [Figure 8] 10 is a flowchart illustrating a process flow in which the lodging estimation system 1 selects an estimation model. [Figure 9] 10 is a flowchart illustrating the flow of processing by the lodging estimation system 1 to estimate the lodging state. DETAILED DESCRIPTION OF THE INVENTION

[0010] A lodging estimation system according to an embodiment of the present invention will be described below with reference to the drawings.

[0011] FIG. 1 is a schematic block diagram showing the configuration of a falling estimation system 1 according to one embodiment of the present invention. The lodging estimation system 1 includes a field survey data acquisition unit 101, a field survey data storage unit 102, an acquired image data acquisition unit 103, a vegetation index calculation unit 104, a vegetation index information storage unit 105, a field survey point data extraction unit 106, a learning model generation unit 107, a model information storage unit 108, a field survey point data extraction unit 109, an estimation accuracy calculation unit 110, an estimation model selection unit 111, a lodging estimation unit 120, a lodging map generation unit 121, and a lodging map output unit 122.

[0012] The lodging estimation system 1 has three functions: learning, verification, and estimation. The learning function generates a learning model based on the correspondence between vegetation indices and lodging states. The verification function verifies the learning model and selects an estimation model that estimates the lodging state based on the verification results. The estimation function estimates the lodging state of plants in a field using the estimation model. In the lodging estimation system 1, the learning function is realized using a field survey data acquisition unit 101, an acquired image data acquisition unit 103, a vegetation index calculation unit 104, a field survey point data extraction unit 106, and a learning model generation unit 107. The verification function is realized using the field survey data acquisition unit 101, the acquired image data acquisition unit 103, the vegetation index calculation unit 104, the estimation accuracy calculation unit 110, and the estimation model selection unit 111. The estimation function is realized using the acquired image data acquisition unit 103, the vegetation index calculation unit 104, a lodging estimation unit 120, a lodging map generation unit 121, and a lodging map output unit 122.

[0013] The field survey data acquisition unit 101 acquires field survey data. The field survey data is a lodging index value obtained as a result of a field survey of the lodging state of plants grown in a farm field. For example, the lodging index value may be a binary index indicating whether or not a plant has lodged, or an index indicating the lodging state at multiple levels. For example, the lodging state may be indicated at multiple levels based on the degree of inclination between the ground surface on which the plant is planted and the stem (culm) of the plant. The lodging state may also be indicated by whether the stem is bent or curved. When conducting a field survey, the person in charge of the survey actually goes to the field and investigates the lodging condition at multiple locations in the field, thereby obtaining lodging index values ​​per unit area at each of multiple different locations. The field survey data acquisition unit 101 acquires lodging index values, which are the results of field surveys conducted based on the lodging state of plants grown in a field at times corresponding to the times when the acquired image data described below was measured. The field survey data acquisition unit 101 writes and stores the acquired field survey data in the field survey data storage unit 102.

[0014] FIG. 2 is a diagram showing an example of field survey data. FIG. 2 shows an example of the field survey data stored in the field survey data storage unit 102. The field survey data is data in which a position ID (Identification), a lodging state, latitude, and longitude are associated with each other. The position ID is an identifier that identifies a position in the field. The position here indicates which of the divided areas into which the field is divided is the divided area. The lodging state indicates the lodging state of the plant at the position (divided area) indicated by the position ID. The latitude indicates the latitude of a predetermined position in the divided area in which the number of stems was counted in the field. The longitude indicates the longitude of a predetermined position in the divided area in which the number of stems was counted in the field. The predetermined position here may be the center of the divided area (e.g., the center of gravity) or a predetermined position on the periphery of the divided area. If the divided area is rectangular, the predetermined position may be, for example, one of the vertices such as the top left vertex or the bottom right vertex, or the center position of one of the four sides.

[0015] Returning to the explanation of FIG. 1, the acquired image data acquisition unit 103 acquires acquired image data obtained by photographing the farm field. The acquired image data is, for example, data in which the luminance value of each pixel is expressed by RGB values ​​(red, green, blue values) and near-infrared light values ​​(NIR values). More specifically, the acquired image data is data including latitude and longitude, and luminance values ​​of red light, blue light, green light, and near-infrared light. The wavelengths of light and their luminance values ​​in the acquired image data are not limited to the luminance values ​​of red light, blue light, green light, and near-infrared light, and may include wavelengths of light and their luminance values ​​necessary for calculating vegetation indices other than the vegetation indices shown in FIG. 3. In this embodiment, the acquired image data may be data obtained by measuring the intensity of each wavelength using a sensor that measures a plurality of wavelengths. It is desirable that the measurement date of the acquired image data and the survey date of the field survey data are the same, but it is sufficient that at least a correlation can be established between the two data, and the date on which the acquired image data was measured and the date on which the field survey was conducted to obtain the field survey data may be different. This is because even if the measurement date of the acquired image data and the survey date of the field survey data are different, the tendency of the lodging state of plants in the field is unlikely to change significantly to the extent that it would affect the estimation of lodging. The acquired image data is data measured by a sensor from above the field. For example, the acquired image data is image data obtained by capturing an image of an area including the field from above. This image may be captured by a person using a sensor, or by a sensor installed or mounted on a structure such as a pylon in or near the field, or on a drone, airplane, or satellite. The field captured here is the same field as the field from which field survey data is acquired, and is a field in which certain plants are growing at the same time as the field survey. The acquired image data acquisition unit 103 outputs the acquired image data to the vegetation index calculation unit 104.

[0016] The vegetation index calculation unit 104 calculates a vegetation index from the acquired image data. There are various types of vegetation indices, such as NDVI (Normalized Difference Vegetation Index), NDRE (Normalized Red Edge Index), and NDWI (Normalized Water Index). NDVI is a normalized value based on the difference in reflectance between near-infrared light and red light. Plants reflect near-infrared wavelengths, but absorb red wavelengths, which are necessary for photosynthesis. Therefore, by obtaining NDVI based on the difference in reflectance between near-infrared light and red light obtained by photographing plants, it can be said that the higher this value, the greater the area of ​​the plant's leaves and stems and the amount of chlorophyll they contain. NDRE is a value obtained by emphasizing the difference in relative intensity between red edge wavelength light and near-infrared light. Like NDVI, NDRE is a vegetation index that correlates with the amount of chlorophyll in leaves, but compared to NDVI, it is used as an indicator of plant health in the later stages of growth, or to detect problems with plant growth earlier than NDVI. There are two types of NDWI: one that defines the amount of water contained in vegetation, and one that defines the amount of water contained in the earth's surface. Of these, the NDWI that defines the amount of water contained in vegetation is a value normalized based on the difference in reflectance between near-infrared light and short-wavelength infrared light, and is an index related to the amount of water contained in plants. The NDWI that defines the amount of water contained in vegetation may also use a value normalized based on the difference in reflectance between green light and short-wavelength infrared light. NDWI, which defines the amount of water contained in the earth's surface, is a value normalized based on the difference in reflectance between red light and short-wavelength infrared light, and is an index of the amount of water contained in the soil. NDWI, which defines the amount of water contained in the earth's surface, may also be a value normalized based on the difference in reflectance between green light and near-infrared light. There is also the IRVI (Improved Ratio Vegetation Index), a vegetation index that improves on the NDVI by reducing the influence of the atmosphere. There are also RGBVI (RGB Vegetation Index) and VARI (Visible Atmospherically Resistant Index), which are vegetation indices based on the RGB values ​​of visible light that does not include near-infrared light, i.e., the luminance values ​​of red light, blue light, and green light.

[0017] Here, it has been discovered that, at a certain time during the plant growth period, a particular vegetation index among several vegetation indices, for example, NDVI, has a particularly strong correlation with field survey data, at another time, NDRE has a particularly strong correlation with field survey data, and at another time, there is a strong correlation between both NDVI and NDRE and field survey data, and so on. This means that the vegetation indices correlated with field survey data vary depending on the time period.

[0018] Here, factors that cause plants to lodge can be thought of as the growth state of the plant (internal factors) and external factors. Internal factors include, for example, thin culms or culm wall thickness, long and thin lower internodes, or withered leaf sheaths, which can cause the plant stem to break easily and lead to lodging. Long culms and many ears can also cause the stem to bend easily and lead to lodging. These internal factors are influenced by the plant's varietal characteristics, the properties of the soil in which the plant is planted (e.g., excess nitrogen), the cultivation environment such as dense planting, and the growth environment such as insufficient sunlight and the occurrence of pests and diseases. External factors include external forces such as wind, rain, and dew that can cause the plant to lodge.

[0019] In this embodiment, the vegetation index calculation unit 104 calculates each of a plurality of types of vegetation indices from the acquired image data, thereby enabling the generation of an estimation model based on a vegetation index that correlates with any of the various factors that may cause lodging. By using such an estimation model to estimate the lodging state, even if there are many factors that could cause lodging, it is possible to use a vegetation index that is correlated with the lodging state obtained from field survey data at the time of year for which lodging is being estimated, making it possible to estimate the lodging state with high accuracy. Furthermore, by calculating multiple types of vegetation indices, it is possible to generate an estimation model based on a vegetation indices that correlate with the color of the plant culm. When a plant falls, the culm can be observed from above, and it is thought that an estimation model based on a vegetation indices that reflect the color of the plant culm can more accurately estimate the state of lodging.

[0020] The vegetation index calculation unit 104 calculates a vegetation index by substituting pixel values ​​of the acquired image data into a vegetation index calculation formula for calculating a vegetation index. The vegetation index calculation unit 104 acquires the vegetation index calculation formula stored in the vegetation index information storage unit 105, for example, by referring to the vegetation index information storage unit 105. The vegetation index calculation unit 104 generates vegetation index image data based on the calculated vegetation index. The vegetation index image data is image data of a vegetation index image in which vegetation indices are associated with pixels.

[0021] Here, the vegetation index calculation unit 104 may perform filtering on the vegetation index image data. Alternatively, the vegetation index calculation unit 104 may perform filtering on the acquired image data. The vegetation index calculation unit 104 may perform filtering on each divided region (described later) or on the entire image. The purpose of filtering is, for example, to remove information unnecessary for estimating the lodging state. More specifically, the purpose of filtering is to remove areas of the image where no vegetation grows so that the areas where no vegetation grows are not unnecessarily emphasized, or to lower the resolution if the resolution is too high. The filter used here may be a predetermined filter, or at least one of a Gaussian filter, a median filter, a band-elimination filter, etc., or may be adaptively determined using a convolutional neural network that simultaneously estimates an optimal filter and an estimation model.

[0022] The vegetation index information storage unit 105 stores vegetation index calculation formulas. FIG. 3 is a diagram showing an example of information stored in the vegetation index information storage unit 105. The vegetation index information storage unit 105 stores information corresponding to each item of a vegetation index ID and a vegetation index calculation formula. The vegetation index ID is an identifier for identifying a new vegetation index. The vegetation index calculation formula is a formula for calculating a vegetation index. In this diagram, a vegetation index calculation formula corresponding to NDVI as a vegetation index identified by a vegetation index ID (VI001) is shown. Also, a vegetation index calculation formula corresponding to IRVI as a vegetation index identified by a vegetation index ID (VI002) is shown.

[0023] While the above description assumes that the field imaged to obtain the acquired image data is the same field as the field where the field survey is conducted, this is not limiting. It is sufficient that there is a correlation in the lodging state between the field survey data and the field where the field survey data is obtained. Specifically, even if the field imaged to obtain the acquired image data and the field where the field survey is conducted are different fields, if the varietal characteristics of the plants growing in the two fields are the same or similar in characteristics that could cause lodging (e.g., culm thickness, culm wall thickness, lower internode length, etc.), there is a possibility that the lodging states in the two fields are correlated. Furthermore, even if the fields are different, such as in neighboring areas, there is a possibility that there is a correlation in the lodging state if the growing environment, such as sunlight conditions, and weather conditions, such as wind, rain, and dew, are similar. In this way, if the lodging conditions of the fields are likely to be correlated, the field imaged to obtain the acquired image data and the field where the field survey is conducted may be different fields.

[0024] FIG. 4 shows an example of a vegetation index image G based on a vegetation index calculated from acquired image data. This vegetation index image G represents the field to be measured, with the vertical direction representing latitude and the horizontal direction representing longitude. This vegetation index image G is vegetation index image data that shows the vegetation index at each position using a color corresponding to the vegetation index calculated based on the acquired image data. Here, positions P1 and P2 are represented by roughly the same color. This color indicates a small vegetation index value. Positions P3 and P4 are represented by roughly the same color. The colors of positions P3 and P4 are different from the colors of positions P1 and P2. The vegetation index values ​​of positions P3 and P4 are larger than the vegetation index values ​​of positions P1 and P2.

[0025] Returning to the explanation of Figure 1, the field survey point data extraction unit 106 extracts, from the vegetation indices for each position in the field calculated by the vegetation index calculation unit 104, a vegetation index that corresponds to a position (divided area) included in the field survey data obtained by the field survey data acquisition unit 101. For example, the field survey point data extraction unit 106 cuts out a divided area in the field where a field survey was conducted from among the divided areas for which a vegetation index was calculated by the vegetation index calculation unit 104. As a result, from the vegetation indices for each position in the field calculated by the vegetation index calculation unit 104, a vegetation index for a position corresponding to the position where the field survey data was obtained is extracted.

[0026] The learning model generation unit 107 generates a learning model. The learning model here is a model that learns the correspondence between the vegetation index and the lodging state using the vegetation index at a position corresponding to the position where the field survey data was obtained and the lodging index value indicating the lodging state in the field survey data. Among the learning models, a model selected by the estimation model selection unit 111 (described later) becomes the estimation model. The estimation model is a model that accurately estimates the lodging state of plants from the vegetation index. The learning model generation unit 107 generates the learning model by, for example, having the learning model learn the relationship between the vegetation index for each position and the lodging index value indicating the lodging state in the field survey data. Examples of the learning model that can be used include well-known algorithms such as neural networks, support vector machines, logistic regression, random forests, and linear regression. The learning model generation unit 107 may also generate a function indicating the relationship between the vegetation index and the growth index value as the learning model by applying regression analysis to the relationship between the vegetation index and the lodging index value obtained from the field survey data. For the regression analysis, linear regression, polynomial regression, Bayesian linear regression, Bayesian neural network, or the like may be used. The learning model generation unit 107 stores information such as the configuration of the generated estimation model in the model information storage unit 108.

[0027] The model information storage unit 108 stores information about the models. The models here are learning models generated by the learning model generation unit 107, and include estimation models selected by the estimation model selection unit 111, which will be described later. FIG. 5 is a diagram showing an example of information stored in the model information storage unit 108. The model information storage unit 108 stores information corresponding to each item, for example, a model ID, a vegetation index ID, a model configuration, and an estimation accuracy. The model ID is an identifier for identifying a model. The vegetation index ID is an identifier for the vegetation index used to generate the estimation model identified by the model ID. The model configuration is the configuration of the model generated by the learning model generation unit 107, such as a polynomial. The estimation accuracy is the estimation accuracy calculated by the estimation accuracy calculation unit 110, which will be described later.

[0028] The field survey point data extraction unit 109 extracts vegetation indices corresponding to positions (divided areas) included in the field survey data as data to be used in calculating the estimation accuracy of the learning model. The method by which the field survey point data extraction unit 109 extracts vegetation indices is the same as the method by which the field survey point data extraction unit 106 extracts vegetation indices, and therefore a description thereof will be omitted. The estimation accuracy calculation unit 110 calculates the estimation accuracy of the learning model generated by the learning model generation unit 107. The estimation accuracy calculation unit 110 calculates the estimation accuracy by comparing the estimation result obtained by inputting the vegetation index extracted by the field survey point data extraction unit 109 (the vegetation index at the position in the field where the lodging state has been field surveyed) into the learning model with the field survey results. The field survey results used for verification here can be field survey data acquired by the field survey data acquisition unit 101 that was not used when generating the learning model but was obtained the following year or later for the same type of crop used to generate the learning model, or field survey data acquired within a certain period before harvesting the same type of crop in the same year. The vegetation index used for verification here can be a vegetation index calculated by the vegetation index calculation unit 104 from the pixel values ​​of pixels corresponding to the field surveyed positions in the acquired image data of the field corresponding to the field survey data used for verification. For example, the estimation accuracy calculation unit 110 calculates the error between the result of estimation of the lodging state using the learning model and the lodging state obtained from field survey data, and calculates statistical quantities of the calculated error, such as the mean value, maximum value, minimum value, mode, variance, standard deviation, etc., as the estimation accuracy. The estimation accuracy calculation unit 110 stores the calculated estimation accuracy in the model information storage unit 108.

[0029] The estimation model selection unit 111 selects an estimation model. For example, the estimation model selection unit 111 determines a learning model whose estimation accuracy is equal to or greater than a threshold value as the estimation model. Alternatively, the estimation model selection unit 111 may select a learning model with the highest estimation accuracy as the estimation model. Furthermore, the estimation model selection unit 111 may store the selection result in the model information storage unit 108.

[0030] The lodging estimation unit 120 estimates the lodging state of plants in a field using an estimation model. The lodging estimation unit 120 acquires a vegetation index for each pixel calculated from the acquired image data of the estimation target. The vegetation index used for the estimation can be the vegetation index calculated by the vegetation index calculation unit 104 from the pixel values ​​of each pixel in the acquired image data of the estimation target. The lodging estimation unit 120 reads the model configuration of the estimation model from the model information storage unit 108 based on an estimation model whose estimation accuracy obtained by the estimation accuracy calculation unit 110 is equal to or greater than a threshold value, or an estimation model selected by the estimation model selection unit 111, and constructs an estimation model based on the read model configuration. The lodging estimation unit 120 inputs the vegetation index into the estimation model to estimate the lodging state corresponding to the vegetation index. The lodging estimation unit 120 outputs the estimation result (lodging index value) to the lodging map generation unit 121.

[0031] The lodging map generation unit 121 generates a lodging map. The lodging map is a map that indicates, for each position in the field, the lodging state (an estimated value of the lodging state) of plants growing at that position. The lodging map generation unit 121 generates the lodging map by associating the lodging state estimated by the lodging estimation unit 120 with each position in the field. The lodging map output unit 122 outputs the lodging map. For example, the lodging map output unit 122 displays the lodging map on a display (not shown) connected to the lodging estimation system 1. This display may be a terminal device such as a user's smartphone connected to the lodging estimation system 1 via the Internet or the like.

[0032] FIG. 6 is a diagram showing an example of a lodging map. This lodging map M shows the growth state of plants growing in a field corresponding to the vegetation index image G shown in FIG. 4. The vertical direction of the lodging map M shown in FIG. 6 is latitude and the horizontal direction is longitude. In the lodging map M, plants growing in colored areas are lodged, and plants growing in uncolored areas are not lodged. In this figure, plants growing at positions P1 and P2 are not lodged. Plants growing at positions P3 and P4 are lodged. In this figure, the state of lodging is indicated by a binary value of "lodged" or "not lodged," but this is not limiting. The lodging map may indicate the state of lodging in multiple stages. In this case, the lodging map M is, for example, a map in which multiple stages of the lodging state are represented by colors corresponding to each stage.

[0033] The functional units of the above-described lodging estimation system 1 (including the field survey data acquisition unit 101, acquired image data acquisition unit 103, vegetation index calculation unit 104, field survey point data extraction unit 106, learning model generation unit 107, field survey point data extraction unit 109, estimation accuracy calculation unit 110, estimation model selection unit 111, lodging estimation unit 120, lodging map generation unit 121, and lodging map output unit 122) may be configured as a processing unit such as a CPU (central processing unit) or a dedicated electronic circuit. Furthermore, the functions of the lodging estimation system 1 may be implemented as a single device by being installed on a single computer, or may be implemented as a system available on the cloud by being installed on a server device connected to a network such as the Internet. The memory units (field survey data memory unit 102, vegetation index information memory unit 105, model information memory unit 108) of the lodging estimation system 1 are configured with storage media such as a hard disk drive (HDD), flash memory, electrically erasable programmable read-only memory (EEPROM), random access read / write memory (RAM), read-only memory (ROM), or any combination of these storage media. Nonvolatile memory may be used as the memory units. Data acquired from outside, values ​​calculated in each unit, and data prepared in advance may also be stored in the memory units.

[0034] Next, the flow of processing performed by the falling estimation system 1 in the above-described embodiment will be described with reference to Fig. 7 to Fig. 9. Fig. 7 to Fig. 9 are flowcharts illustrating the flow of processing performed by the falling estimation system 1.

[0035] Figure 7 shows the flow of processing by which the lodging estimation system 1 generates a learning model. The field survey data acquisition unit 101 of the lodging estimation system 1 acquires field survey data for learning (step S10). To generate a learning model, a field survey of the field is conducted and the field is measured using sensors. The field survey data acquisition unit 101 acquires lodging index values ​​based on the results of the field survey as field survey data. The results of the field survey are acquired based on the details entered by a person in charge of input via an input device such as a keyboard, mouse, or panel. The acquired image data acquisition unit 103 of the lodging estimation system 1 acquires acquired image data for learning (step S11). The acquired image data acquisition unit 103 acquires, as acquired image data, images of the field where the field survey was conducted in step S10 at a time corresponding to the time when the field survey was conducted in step S10. The vegetation index calculation unit 104 of the lodging estimation system 1 calculates a vegetation index based on pixel values ​​of the acquired image data acquired in step S11 (step S12). Here, the vegetation index calculation unit 104 calculates, from the acquired image data, each of multiple types of vegetation indices for which a learning model is to be generated. The field survey point data extraction unit 106 extracts the vegetation indices calculated in step S12 that correspond to each location (divided area) where the field survey was conducted in step S10, and generates learning data that associates the vegetation indices with the lodging state (e.g., whether or not the plant is lodged) (step S13). The learning model generation unit 107 generates a learning model that has learned the correspondence between the learning data generated in step S13, i.e., the vegetation index and the lodging state (step S14). The learning model generation unit 107 stores information about the generated learning model in the model information storage unit 108 (step S15). The learning model generation unit 107 determines whether or not a learning model corresponding to each vegetation indices has been generated for all of the target vegetation indices (step S16), and if there are any vegetation indices for which a learning model has not been generated, the process returns to step S13 and a learning model corresponding to each vegetation indices is generated. On the other hand, if a learning model corresponding to each vegetation indices has been generated for all of the target vegetation indices in step S16, the lodging estimation system 1 ends the process. The order of steps S10 and S11 may be interchanged.

[0036] 8 shows the flow of processing in which the lodging estimation system 1 verifies a learning model and selects an estimation model based on the verification results. Steps S20 to S21 are the same as steps S10 to S11 except that "for learning" is changed to "for verification." The vegetation index calculation unit 104 of the lodging estimation system 1 calculates a vegetation index corresponding to the learning model to be verified based on the pixel values ​​of the acquired image data acquired in step S21 (step S22). Here, the vegetation index calculation unit 104 calculates a vegetation index corresponding to the learning model to be verified from the acquired image data. The field survey point data extraction unit 109 extracts the vegetation indices calculated in step S22 that correspond to each location (divided area) where the field survey was conducted in step S20, and inputs the vegetation indices into the learning data model (step S23). The estimation accuracy calculation unit 110 of the lodging estimation system 1 inputs the vegetation index calculated in step S22 corresponding to the learning model to be verified into the learning model (step S23). The estimation accuracy calculation unit 110 verifies the estimation result by comparing the estimation result of the lodging state output from the learning model with the lodging state in the implementation survey data (step S24). The estimation accuracy calculation unit 110 performs the processes shown in steps S22 to S24 for each pixel corresponding to each position of the field in the acquired image data for verification acquired in step S21. The estimation accuracy calculation unit 110 calculates the estimation accuracy of the learning model. The estimation model selection unit 111 determines whether the estimation accuracy of the learning model is the highest value (step S25). Alternatively, in step S25, the estimation model selection unit 111 may perform threshold judgment to determine whether the estimation accuracy of the learning model is equal to or greater than a threshold value. If the estimation accuracy of the learning model is the highest value, the estimation accuracy calculation unit 110 selects the learning model as the estimation model (step S26). Alternatively, when threshold determination is performed in step S25, if the estimation accuracy of the learning model is equal to or greater than the threshold, the estimation accuracy calculation unit 110 designates the learning model as the estimation model. On the other hand, if the estimation accuracy of the learning model is not the highest value, the estimation accuracy calculation unit 110 does not designate the learning model as the estimation model. Alternatively, when threshold determination is performed in step S25, if the estimation accuracy of the learning model is less than the threshold, the estimation accuracy calculation unit 110 does not designate the learning model as the estimation model. The estimation accuracy calculation unit 110 may store the determination result of whether or not to designate the learning model as the estimation model in the model information storage unit 108. The estimation model selection unit 111 determines whether all of the learning models to be verified have been verified (step S27), and if there are any learning models that have not been verified, the process returns to step S22 and verifies those learning models. On the other hand, if verification has been completed for all of the learning models to be verified in step S27, the falling estimation system 1 ends the process. In addition, the estimation accuracy calculation unit 110 may use a portion of the learning data generated in step S13 (data that associates the vegetation index with the state of lodging as a result of the field survey) for learning, and the remaining acquired image data for verification.

[0037] FIG. 9 shows a flow of processing in which the lodging estimation system 1 estimates the lodging state using an estimation model. The acquired image data acquisition unit 103 of the lodging estimation system 1 acquires acquired image data of the estimation target (step S30). The vegetation index calculation unit 104 calculates a vegetation index corresponding to the estimation model from the acquired image data acquired in step S30 (step S31). The lodging estimation unit 120 inputs the vegetation index calculated in step S31 into the estimation model (step S32). The lodging estimation unit 120 acquires the estimation result of the lodging state output from the estimation model (step S33). The lodging map generation unit 121 generates a lodging map (step S34). The lodging map generation unit 121 generates a lodging map by associating the estimation result of the lodging state estimated by the lodging estimation unit 120 with each pixel corresponding to the position in the field. The lodging map output unit 122 outputs the lodging map generated by the lodging map generation unit 121 (step S35).

[0038] As described above, the lodging estimation system 1 of this embodiment includes an acquired image data acquisition unit 103, a field inspection data acquisition unit 101, and a lodging estimation unit 120. The acquired image data acquisition unit 103 is an example of an image acquisition unit and acquires acquired image data by photographing the field. The field inspection data acquisition unit 101 acquires lodging index values, which are the results of field inspections of the lodging state of plants grown in the field, at times corresponding to the time the acquired image data was measured. The lodging estimation unit 120 is an example of an estimation unit and estimates a lodging index value based on acquired image data of the estimation target field measured based on the relationship between the acquired image data and the actual lodging index value obtained through the field inspection. This allows the lodging estimation system 1 of this embodiment to grasp the correspondence between photographed images of the field and the results of the field inspection, making it possible to estimate the lodging state in areas of the field that have not been inspected based on this correspondence. Therefore, even in large-scale farm fields, the lodging state of plants growing in the field can be easily estimated without increasing personnel costs.

[0039] The lodging estimation system 1 of this embodiment also includes a vegetation index calculation unit 104 and a learning model generation unit 107. The vegetation index calculation unit 104 calculates a vegetation index from the acquired image data for each field position corresponding to the lodging index value obtained as a result of field surveys. The learning model generation unit 107 creates a learning model that learns the relationship between the lodging index value obtained as a result of field surveys and the vegetation index obtained from the acquired image data for each field position. The lodging estimation unit 120 inputs the acquired image data of the estimation target into the learning model and estimates the lodging index value as a lodging index value based on the acquired image data of the estimation target. This allows the lodging estimation system 1 of this embodiment to easily estimate the lodging state using machine learning techniques.

[0040] Furthermore, in the lodging estimation system 1 of this embodiment, the vegetation index calculation unit 104 calculates multiple types of vegetation indices from the acquired image data. The learning model generation unit 107 creates multiple learning models corresponding to the multiple types of vegetation indices calculated by the vegetation index calculation unit 104. The lodging estimation unit 120 inputs the estimation target acquired image data into an estimation model selected by the estimation model selection unit 111 from the multiple learning models generated by the learning model generation unit 107, based on the estimation accuracy calculated by the estimation accuracy calculation unit 110 during verification, and selects an estimation model with an estimation accuracy equal to or higher than a threshold value or with the highest estimation accuracy. The lodging estimation unit 120 estimates the lodging index value based on the estimation target acquired image data. The lodging estimation unit 120 uses the learning model selected by the estimation model selection unit 111 from the multiple learning models generated by the learning model generation unit 107 as the estimation model. The estimation model selection unit 111 calculates the estimation accuracy of each of the learning models calculated by the estimation accuracy calculation unit 110. As a result, the lodging estimation system 1 of the embodiment can estimate the lodging state using a learning model that is particularly correlated with the lodging state, that is, a learning model that can estimate the lodging state with high accuracy.

[0041] (Modification of the embodiment) A modified example of the embodiment will now be described. In this modified example, a calculation formula for calculating a new vegetation index is generated by modifying an existing calculation formula for calculating a vegetation index. The existing vegetation index here is a vegetation index stored in the vegetation index information storage unit 105. By calculating a new vegetation index, it becomes possible to search for a vegetation index that has a higher degree of correlation with the lodging state of a plant. In particular, if a vegetation index that has a higher degree of correlation with the color of the plant's culm observed from above when a plant has lodged can be generated, it becomes possible to generate a learning model that can accurately estimate the lodging state regardless of the plant's growth condition. Specifically, the vegetation index calculation unit 104 changes the combination of light components in the calculation formula for determining the vegetation index, and changes a wavelength band corresponding to a specific color to a wavelength band corresponding to another color.

[0042] For example, when the base vegetation index is RGBVI, the calculation formula for determining the vegetation index can be expressed as formula (1): RGBVI in formula (1) is the vegetation index. The vegetation index calculation unit 104 changes the combination of light components in the RGBVI calculation formula shown in formula (1) to generate a calculation formula for a new vegetation index, for example, a combination of blue light and green light. In this case, the calculation formula for the new vegetation index can be expressed as formula (2). RGBVI# in formula (2) is the new vegetation index. In equations (1) and (2), Red is the light component of red light, Blue is the light component of blue light, and Green is the light component of green light.

[0043] RGBVI={Green^2-(Red×Blue)} / {Green^2+(Red×Blue)} ...Equation (1) RGBVI#={Red^2-(Blue×Green)} / {Red^2+(Blue×Green)} …Equation (2)

[0044] As described above, in the lodging estimation system 1 according to the modified embodiment, the vegetation index calculation unit 104 calculates a new vegetation index. The new vegetation index is a vegetation index calculated using a new formula for calculating a vegetation index that is generated by modifying an existing formula for calculating a vegetation index. For example, the vegetation index calculation unit 104 calculates the new vegetation index using a formula in which the combination of light components in the formula for calculating a vegetation index is changed and a wavelength band corresponding to a specific color is replaced with a wavelength band corresponding to another color. The learning model generation unit 107 creates a learning model corresponding to the new vegetation index calculated by the vegetation index calculation unit 104. The lodging estimation unit 120 inputs the estimation target acquired image data into the learning model generated by the learning model generation unit 107 and estimates the lodging index value based on the estimation target acquired image data. As a result, the lodging estimation system 1 according to the modified embodiment can identify a new vegetation index that has a higher degree of correlation with the lodging state of an object, and estimate the lodging state using a learning model corresponding to the identified new vegetation index, thereby making it possible to accurately estimate the lodging state.

[0045] In the above-described embodiment, the case of estimating the lodging state of wheat at harvest time has been described as an example, but the present invention is not limited to this and can be applied to estimating the lodging state at any time during the plant growth period. It can also be applied to estimating the lodging state of plants other than wheat. Examples of other plants include rice, barley, and rye.

[0046] Each component of the fallen tree estimation system in the above-described embodiment may be implemented by a computer. In this case, a program for implementing the functions may be recorded on a computer-readable recording medium, and the program may be loaded into a computer system and executed. The term "computer system" as used herein includes hardware such as an operating system (OS) and peripheral devices. The term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. The term "computer-readable recording medium" may also include devices that dynamically store programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or telephone lines, or devices that store programs for a fixed period of time, such as volatile memory within a computer system serving as a server or client. The program may be for implementing only a portion of the functions described above, or may be capable of implementing the functions in combination with a program already stored in the computer system. The program may also be implemented using a programmable logic device such as a field programmable gate array (FPGA).

[0047] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention. [Explanation of symbols]

[0048] 1...Lodging estimation system, 101...Field survey data acquisition unit, 103...Acquired image data acquisition unit, 104...Vegetation index calculation unit, 106...Field survey point data extraction unit, 107...Learning model generation unit, 109...Field survey point data extraction unit, 110...Estimation accuracy calculation unit, 111...Estimation model selection unit, 120...Lodging estimation unit, 121...Lodging map generation unit, 122...Lodging map output unit

Claims

1. an acquired image data acquisition unit that acquires acquired image data obtained by photographing the farm field; a field survey data acquisition unit that acquires a lodging index value that is a result of a field survey of the lodging state of plants grown in a field at a time corresponding to the time when the acquired image data was measured; a lodging estimation unit that estimates a lodging index value based on the acquired image data of an estimation target obtained by measuring the estimation target field, based on the relationship between the acquired image data and a lodging index value as an actual result obtained by a field survey; A lodging estimation system having the above.

2. a vegetation index calculation unit that calculates a vegetation index from the acquired image data for each position in the field corresponding to a lodging index value as an actual result obtained by a field survey; a learning model generation unit that generates, for each position in the field, a learning model that has learned the relationship between a lodging index value as an actual result obtained by a field survey and a vegetation index obtained from the acquired image data; and the lodging estimation unit estimates an estimation result of the lodging index value obtained by inputting the estimation target acquired image data into the learning model generated by the learning model generation unit, as the lodging index value based on the estimation target acquired image data. The system for estimating lodging according to claim 1 .

3. the vegetation index calculation unit calculates a plurality of types of vegetation indices from the acquired image data, the learning model generation unit creates a plurality of learning models corresponding to the plurality of types of vegetation indices calculated by the vegetation index calculation unit, the lodging estimation unit inputs the acquired image data of the estimation target into an estimation model selected from the plurality of learning models generated by the learning model generation unit according to estimation accuracy, and estimates the result of the estimation of the lodging index value obtained as the lodging index value based on the acquired image data of the estimation target. The system for estimating lodging according to claim 2 .

4. the vegetation index calculation unit calculates a new vegetation index using a calculation formula for calculating a new vegetation index generated by making changes to the calculation formula for calculating the vegetation index; the learning model generation unit creates a learning model corresponding to the new vegetation index calculated by the vegetation index calculation unit, the lodging estimation unit estimates an estimation result of the lodging index value obtained by inputting the estimation target acquired image data into the learning model generated by the learning model generation unit, as the lodging index value based on the estimation target acquired image data. The system for estimating lodging according to claim 2 .

5. A lodging estimation method performed by a computer for use in a lodging estimation system, comprising: an acquired image data acquisition unit acquires acquired image data obtained by photographing the farm field; obtaining a lodging index value as a result of an on-site survey of the lodging state of plants grown in a field at a time corresponding to the time when the obtained image data was measured; a lodging estimation unit that estimates a lodging index value based on the acquired image data of the estimation target obtained by measuring the estimation target field, based on a relationship between the acquired image data and a lodging index value as an actual result obtained by a field survey; Lodging estimation method.

Images