Prediction method, prediction program, system and model generation method for optimum midseason drainage date for rice cultivation, and rice cultivation method

Abstract

To provide a technology for supplying information regarding an optimum midseason drainage date in rice cultivation.SOLUTION: A method for predicting an optimum midseason drainage date for rice cultivation in a paddy field includes steps, by a computer, of: acquiring an image of the paddy field; converting the image into a vegetation coverage rate in the paddy field; and predicting a remaining accumulated temperature in the paddy field based on the vegetation coverage rate. The remaining accumulated temperature represents the accumulated temperature required until the optimum midseason drainage date. The method further includes a step, by the computer, referring to weather information corresponding to the paddy field so as to generate a prediction result of the optimum midseason drainage date based on a period from the time the image is acquired until the accumulated temperature reaches the remaining accumulated temperature.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to predictions regarding rice cultivation, and in particular to predictions of suitable days for mid-season drying of rice in fields. [Background technology]

[0002] Various studies have been conducted on the use of IT technology in the cultivation of agricultural crops in farm fields. For example, Japanese Patent Laid-Open Publication No. 2017-046639 (Patent Document 1) discloses a technology for calculating the amount of fertilizer to be applied to the crops in a farm field by using images of the farm field. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2017-046639 Summary of the Invention [Problem to be solved by the invention]

[0004] In addition to fertilizer application rates, much other information is required for the cultivation of agricultural crops in fields. For example, the cultivation of paddy rice involves a process called "mid-season drying" as an example of water management, and carrying out this process at the right time is important from the perspectives of preventing the generation of harmful gases, suppressing excessive tillers, and improving rice quality and yield.

[0005] In the past, skilled workers would patrol the fields, visually observe the growth conditions of the rice plants, and identify the best day for carrying out the mid-drying process (hereinafter also referred to as the "best day for mid-drying") based on their experience and knowledge. Because the timing of carrying out the mid-drying process is important as described above, there is a need to provide information to inexperienced workers regarding the identification of the best day for mid-drying.

[0006] The present disclosure has been devised in consideration of the above-mentioned circumstances, and its purpose is to provide a technique for predicting the optimum date for mid-season drying in rice cultivation. [Means for solving the problem]

[0007] The present invention provides the following prediction method, rice cultivation method, prediction program, system, and model generation method.

[0008] [1] A method for predicting a suitable date for mid-season drying of paddy rice in a field, comprising: A computer acquires an image of the field; The computer converts the image into a vegetation cover in the field; and a step of the computer predicting a residual accumulated temperature in the field based on the vegetation cover ratio, The remaining accumulated temperature represents the accumulated temperature required until the suitable day for mid-drying, A prediction method comprising a step in which the computer generates a prediction result of the suitable day for mid-drying based on the period from the time the image was acquired until the accumulated temperature reaches the remaining accumulated temperature by referring to weather information corresponding to the field.

[0009] [2] The step of predicting the residual accumulated temperature includes applying the image to a prediction model; The prediction model is configured to output a prediction result of the residual accumulated temperature by inputting a vegetation cover ratio through machine learning processing using training data, The prediction method described in [1], wherein each of the two or more datasets included in the training data includes vegetation cover tagged with residual accumulated temperature.

[0010] [3] a step in which the computer acquires the variety of paddy rice in the field; The prediction method described in [2], further comprising a step in which the computer selects a model corresponding to the variety from two or more models as the prediction model.

[0011] [4] The computer acquires location information of the field; The prediction method according to any one of [1] to [3], further comprising a step of acquiring weather information corresponding to the location information as weather information corresponding to the field.

[0012] [5] A step in which the terminal takes a photograph of the image; the terminal transmitting the image to the computer; The prediction method according to [1], wherein in the step of capturing an image, information regarding the angle at which the terminal captures the image is output at the terminal.

[0013] [6] A method for cultivating rice using the prediction method according to any one of [1] to [5].

[0014] [7] A prediction program for predicting a suitable date for mid-season drying of paddy rice in a field, comprising: The prediction program is executed by a processor of a computer, thereby causing the computer to: acquiring an image of the field; converting the image into a percent vegetation cover in the field; and predicting a residual accumulated temperature in the field based on the vegetation coverage ratio; The remaining accumulated temperature represents the accumulated temperature required until the suitable day for mid-drying, The prediction program causes the computer to perform a step of generating a prediction result of the suitable day for mid-drying based on the period from the time the image was acquired until the accumulated temperature reaches the remaining accumulated temperature by referring to weather information corresponding to the field.

[0015] [8] A system for predicting the best day for mid-season drying of paddy rice in a field, comprising a terminal and a computer, The terminal a camera that captures an image of the field; a first communication interface for transmitting the image to the computer; The computer a second communication interface for acquiring the image; a controller that converts the image into a vegetation coverage ratio in the field; The controller Predicting a residual accumulated temperature in the field based on the vegetation coverage rate; The residual accumulated temperature is represents the accumulated temperature required until the suitable day for mid-drying, The controller By referring to meteorological information corresponding to the field, a prediction result of the suitable day for mid-season drying is generated based on the period from the time when the image was acquired until the accumulated temperature reaches the remaining accumulated temperature; The system transmits the prediction result to the terminal.

[0016] [9] A method for generating a model for predicting the remaining accumulated temperature, which is the temperature required until the suitable date for mid-season drying of paddy rice in a field, for predicting the suitable date for mid-season drying, comprising: acquiring one or more images of the field; each of the one or more images is associated with an accumulated temperature of the field; A step of identifying a suitable day for mid-season drainage in the field where the one or more images were taken; A step of generating a data set by tagging the vegetation coverage rate of each of the one or more images as the remaining accumulated temperature, which is a value obtained by subtracting the accumulated temperature of each of the one or more images from the total accumulated temperature, which is the accumulated temperature corresponding to a suitable day for mid-season drying in the field; A method for generating a model comprising: using the data set of each of the one or more images as training data to perform machine learning on a model that outputs residual accumulated temperature when vegetation coverage is input.

[0017]

[10] A model generation method described in [9], wherein in the step of identifying the suitable day for mid-drying, the total accumulated temperature is identified based on the number of stems. [Effects of the Invention]

[0018] According to the present invention, a technique for providing information regarding the optimum date for mid-season drying of rice cultivation is provided. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a diagram illustrating an example of an overall configuration of a prediction system. [Figure 2] FIG. 10 is a diagram showing an example of the functional configuration of the server 1 used to predict the best day for mid-drying. [Figure 3] 2 is a diagram illustrating an example of the hardware configuration of a server 1 and a terminal 2. FIG. [Figure 4] FIG. 2 is a diagram for explaining generation and use of a prediction model 123. [Figure 5] FIG. 10 is a diagram showing the change in the number of stalks per unit area determined from an image of a certain field, together with the accumulated temperature in the field. [Figure 6] 10 is a flowchart of an example of processing performed in the server 1 to generate a prediction model 123. [Figure 7] 10 is a flowchart of an example of processing performed in each of the server 1 and the terminal 2 to predict a suitable day for mid-drying. [Figure 8] FIG. 2 is a diagram showing an example of a viewfinder image. [Figure 9] FIG. 10 is a diagram showing images after processing using four types of processing patterns, together with the image before processing. [Figure 10] FIG. 10 is a diagram showing images after processing using four types of processing patterns. [Figure 11] FIG. 2 is a diagram for explaining three types of states in which an image is captured. [Figure 12] FIG. 10 is a diagram for explaining cross-validation without including one field. DETAILED DESCRIPTION OF THE INVENTION

[0020] An embodiment of the present invention will be described below with reference to the drawings. In the following description, the same parts and components are designated by the same reference numerals. Their names and functions are also the same. Therefore, the description thereof will not be repeated.

[0021] [Overall configuration of the prediction system] 1 is a diagram showing an example of the overall configuration of a prediction system. The prediction system uses an image of a farm field to generate a prediction result of a suitable day for mid-season draining of paddy rice in the farm field.

[0022] More specifically, the prediction system includes a server 1. A user 5 uses a terminal 2 to take an image IM of a field and transmit the image IM to the server 1. The server 1 uses the image IM to calculate the vegetation coverage ratio in the image IM, predicts the remaining accumulated temperature using the vegetation coverage ratio, and predicts the suitable day for mid-season drying using the remaining accumulated temperature. The remaining accumulated temperature refers to the accumulated temperature required from a given reference timing to the suitable day for mid-season drying. In this specification, the vegetation coverage ratio refers to the ratio of the pixel area attributable to plants in the image to the pixel area of ​​the entire image.

[0023] [Summary of forecast for optimal days for mid-drying] 2 is a diagram showing an example of the functional configuration of the server 1 used to predict the best day for mid-season drying. In one implementation example, the server 1 has functions shown as an acquisition unit 151, a conversion unit 152, a prediction unit 153, and a prediction unit 154. Each of these functions is realized by the arithmetic processing unit of the controller executing a given application program.

[0024] The server 1 uses a first program 122, a prediction model 123, and a second program 124. The first program 122 calculates the vegetation coverage rate in the field by processing an image of the field. In this specification, calculating the vegetation coverage rate from an image is sometimes referred to as "converting an image to vegetation coverage rate." The prediction model 123 predicts the remaining accumulated temperature from the vegetation coverage rate. When the vegetation coverage rate is calculated from an image, the reference timing for the remaining accumulated temperature obtained as a prediction result for that vegetation coverage rate is the time when the image was captured. The second program 124 uses weather data corresponding to the field to output the number of days remaining until the suitable date for mid-season drying based on the remaining accumulated temperature.

[0025] By executing the above application program, the server 1 performs the following operations.

[0026] The acquisition unit 151 acquires images of the rice field from the terminal 2.

[0027] The conversion unit 152 uses the first program 122 to convert the image acquired by the acquisition unit 151 into a vegetation coverage ratio.

[0028] The prediction unit 153 uses the prediction model 123 to obtain the prediction result of the remaining accumulated temperature from the vegetation coverage ratio output by the conversion unit 152.

[0029] The prediction unit 154 uses the second program to obtain a prediction result of the number of days remaining until the suitable day for mid-drying from the prediction result of the remaining accumulated temperature output by the prediction unit 153.

[0030] The prediction result of the number of remaining days output by the prediction unit 153 and the reference date for the number of remaining days are used to generate and output a prediction result of the suitable day for mid-drying.

[0031] [Hardware configuration] 3 is a diagram showing an example of the hardware configuration of the server 1 and the terminal 2. The hardware configuration of each of the server 1 and the terminal 2 will be described below.

[0032] (Server 1) The server 1 includes a controller 11, a memory 12, a communication interface 13, a display 14, and an input device 15. The server 1 may be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed by software.

[0033] The controller 11 includes an arithmetic processing device such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The controller 11 may include a memory such as a ROM (Read Only Memory) or a RAM (Random Access Memory). The arithmetic processing device of the controller 11 executes various computer programs stored in the ROM or memory 12, and controls the operation of each hardware element in the server 1.

[0034] The memory 12 includes a nonvolatile storage device such as a hard disk or an SSD (Solid State Drive). Various computer programs and data are stored in the memory 12. The memory 12 may be configured with multiple storage devices, or may be an external storage device connected to the server 1.

[0035] In FIG. 3, data 120 is shown as data stored in the memory 12.

[0036] An application program 121 (corresponding to the "application program" described with reference to FIG. 2) is shown as an example of a computer program stored in memory 12. A first program 122 and a second program 124 are also shown as examples of computer programs stored in memory 12.

[0037] A prediction model 123 is shown as data (such as various parameter settings) that defines the model used in the server 1.

[0038] The computer program (computer program product) stored in memory 12 may be provided by a non-transitory recording medium on which the computer program is readably recorded. The recording medium may be a portable memory such as a CD-ROM, a USB memory, or an SD (Secure Digital) card. In this case, controller 11 reads the desired computer program from the recording medium using a reading device (not shown) and stores the read computer program in memory 12. Alternatively, the computer program may be provided via communication. The computer program can be deployed to be executed on multiple computers located on a single computer or at one site, or distributed across multiple sites and interconnected by a communication network.

[0039] The communication interface 13 includes a communication device for performing processing related to communication via the network N. In the prediction system of the present disclosure, the server 1, the terminal 2, and the weather information server 3 are connected via the network N. The controller 11 transmits and receives various types of information to and from the terminal 2 and the weather information server 3 via the communication interface 13.

[0040] The display 14 is realized by a display device such as a liquid crystal panel, an organic EL (Electro Luminescence) display, etc. The display 14 displays various information in accordance with instructions from the controller 11.

[0041] The input device 15 is an interface that accepts user operations, such as a keyboard, a touch panel device with a built-in display, and / or a microphone. The input device 15 accepts operations from the user and sends a control signal to the controller 11 according to the content of the operation.

[0042] (Terminal 2) The terminal 2 includes a controller 21, a memory 22, a communication interface 23, a display 24, an input device 25, a camera 26, a GPS (Global Positioning System) receiver 27, and an acceleration sensor 28. The terminal 2 is a general-purpose information processing terminal, such as a smartphone, a tablet terminal, or a laptop computer.

[0043] The controller 21 includes a processing unit such as a CPU, a GPU, etc. The controller 21 may include a memory such as a ROM or a RAM. The processing unit of the controller 21 executes various computer programs stored in the ROM or memory 22, and controls the operation of each hardware element in the terminal 2.

[0044] The memory 22 includes a nonvolatile storage device such as an embedded Multi Media Card (eMMC) or a Universal Flash Storage (UFS), etc. The memory 22 stores various computer programs (application programs 221) and data (data 220).

[0045] The communication interface 23 includes a communication device for performing processing related to communication via the network N. The controller 21 transmits and receives various types of information to and from the server 1 through the communication interface 23.

[0046] The display 24 is realized by a display device such as a liquid crystal panel, an organic EL display, etc. The display 24 displays various information in accordance with instructions from the controller 21.

[0047] The input device 25 is an interface that accepts user operations, such as a keyboard, a touch panel device with a built-in display, and / or a microphone. The input device 25 accepts operations from the user and sends a control signal to the controller 21 according to the content of the operation.

[0048] Camera 26 generates imaging data in response to an instruction from controller 21, and sends the imaging data to controller 21. In this way, controller 21 acquires imaging data (captured image).

[0049] The GPS receiver 27 receives signals (GPS signals) from GPS satellites and sends the GPS signals to the controller 21. The controller 21 generates position information of the terminal 2 based on the GPS signals.

[0050] The acceleration sensor 28 sends acceleration data in three axial directions to the controller 21. The controller 21 generates angle information of the terminal 2 based on the acceleration data in the three axial directions. The angle information in the terminal 2 may be based on data other than acceleration data. A gyro sensor may be provided in the terminal 2 in addition to or instead of the acceleration sensor 28, and the controller 21 may generate angle information based on data from the gyro sensor.

[0051] [Residual accumulated temperature forecast model] Fig. 4 is a diagram for explaining the generation and use of the prediction model 123. Fig. 4 shows the types of information used in the learning phase and the use phase of the prediction model 123.

[0052] More specifically, in the learning phase, multiple data sets generated for each of multiple training images are used as training data. Each data set has a vegetation cover rate tagged with the residual accumulated temperature. In the server 1, a machine learning model generation program is executed to generate a prediction model 123. The prediction model 123 is, for example, a simple regression model with the vegetation cover rate as an explanatory function and the residual accumulated temperature as a target variable. In the server 1, the application program 121 may have the functionality of the machine learning model generation program, or the machine learning model generation program may be used as a program separate from the application program 121.

[0053] In the utilization phase, the image is converted into a vegetation coverage ratio, and the vegetation coverage ratio is applied to the prediction model 123 to obtain a prediction result of the residual accumulated temperature corresponding to the vegetation coverage ratio.

[0054] [Generating training data] As described above, in each dataset of the training data, the vegetation coverage is tagged with the remaining accumulated temperature. The remaining accumulated temperature is calculated using the total accumulated temperature (the accumulated temperature from the planting date to the suitable day for mid-season drying) corresponding to the field from which each image was taken. More specifically, when the vegetation coverage is determined from a certain image, the accumulated temperature from the planting date to the time the image was taken is calculated. The calculated accumulated temperature is then subtracted from the total accumulated temperature to calculate the remaining accumulated temperature.

[0055] An example of a method for calculating the total accumulated temperature corresponding to each image will now be described with reference to Figure 5. Figure 5 is a diagram showing the change in the number of stalks per unit area determined from an image of a certain field, together with the accumulated temperature in the field.

[0056] In Figure 5, the vertical axis represents the number of stems per unit area (1 square meter), and the horizontal axis represents the accumulated temperature since the planting date. In the example of Figure 5, the number of stems per unit area identified for each of six images taken at different times for a certain field is shown as six measurement points. In the example of Figure 5, two adjacent measurement points are connected by a line.

[0057] To determine the total accumulated temperature, first, for paddy rice, the number of stalks per unit area (for example, 1 tsubo (approximately 3.306 m)) is used as a guideline for mid-season drying. 2 ) is specified. A commonly used value may be used as the number of stems per unit area that is used as a guideline for mid-drying. In Figure 5, the value NV is shown as the number of stems per unit area that is used as a guideline for mid-drying.

[0058] Next, two measurement points adjacent to the number of stems per unit area, which is used as a guideline for mid-season drying, are identified. In Figure 5, points Px and Py are shown as the two measurement points sandwiching the value NV.

[0059] Then, on the line connecting the two measurement points (points Px and Py), the accumulated temperature corresponding to the number of stalks per unit area, which is considered to be a guideline for mid-season drying, is identified as the total accumulated temperature. In Figure 5, the total accumulated temperature TT is shown as the total accumulated temperature identified in this way.

[0060] Then, for each data set, the remaining accumulated temperature is determined.

[0061] More specifically, each image on which each data set of training data is based is associated with the location and date / time of the image. By referencing weather information for the region corresponding to the field of each image, the accumulated temperature in the field from the planting date to the date / time when each image was taken is determined. The accumulated temperature thus determined is subtracted from the total accumulated temperature determined as described above to calculate the remaining accumulated temperature corresponding to each image. Note that, in this specification, weather information includes actually measured weather information and weather forecasts.

[0062] On the other hand, the vegetation cover ratio in each data set is determined by converting the image on which each data set is based into the vegetation cover ratio. As a method for converting an image into the vegetation cover ratio, existing technology (for example, Patent Document 1) can be used.

[0063] [Process flow (predictive model generation)] 6 is a flowchart of an example of processing performed in the server 1 to generate the prediction model 123. In one implementation example, the processing in FIG. 6 is realized in the server 1 by the arithmetic processing device executing a program for generating a machine learning model. In one implementation example, the server 1 starts the processing in FIG. 6 in response to input of an instruction to start the processing in FIG. 6.

[0064] Referring to FIG. 6, in step S10, the server 1 acquires one or more images for generating training data.

[0065] In step S12, the server 1 calculates the total accumulated temperature for the one or more images acquired in step S10.

[0066] In step S14, the server 1 converts each of the one or more images into a vegetation coverage ratio.

[0067] In step S16, the server 1 identifies the remaining accumulated temperature for each vegetation coverage ratio acquired in step S14.

[0068] In step S18, the server 1 generates training data for one or more images. The training data includes one or more data sets. Each data set includes the vegetation coverage and residual accumulated temperature determined for each of the one or more images.

[0069] In step S20, the server 1 performs a learning process for the prediction model using the training data generated in step S18. As a result of the learning process, a prediction model is generated. Thereafter, the server 1 ends the process of FIG. 6.

[0070] According to the processing of FIG. 6 described above, a prediction model is generated that outputs a prediction result of the remaining accumulated temperature by inputting the vegetation cover ratio.

[0071] [Processing flow (prediction of the best day for mid-drying)] Fig. 7 is a flowchart of an example of processing performed in each of the server 1 and the terminal 2 to predict a suitable day for mid-season drying. In one implementation example, in the server 1, the processing in Fig. 7 is realized by the arithmetic processing unit executing the application program 121. In the terminal 2, the processing in Fig. 7 is realized by the arithmetic processing unit executing the application program 221. In one implementation example, the terminal 2 starts the processing in Fig. 7 in response to an operation being performed in the application program 221 to predict a suitable day for mid-season drying.

[0072] In step S30, the terminal 2 notifies the angle of the camera 26. Here, the state of the terminal 2 when step S30 is performed will be described.

[0073] When an operation for predicting a suitable day for mid-season drainage is performed in the application program 221, the application program 121 instructs the user 5 to photograph the field with the camera 26. A viewfinder image of the camera 26 is displayed on the display 24. The user 5 presses the photograph button while looking at the viewfinder image.

[0074] In the application program 221, an appropriate angle of the camera 26 is set when photographing the farm field. The notification in step S30 provides the user 5 with a hint for determining whether the angle of the camera 26 is appropriate. The notification in step S30 may further include information indicating that the camera 26 is at an appropriate angle.

[0075] Fig. 8 is a diagram showing an example of a viewfinder image. Scales are added to the top, bottom, left, and right of the viewfinder image in Fig. 8. The top and bottom scales represent the yaw angle of the camera 26, and the left and right scales represent the pitch angle of the camera 26.

[0076] The viewfinder image is further marked with reference lines 801 and 802. Reference line 801 represents the actual yaw angle of camera 26. Reference line 802 represents the actual pitch angle of camera 26. When the angle of camera 26 is changed, the positions of reference lines 801 and 802 do not change in the viewfinder image, but the numerical values ​​on the up / down and / or left / right scales change. On the up / down scale, the numerical value corresponding to reference line 801 represents the yaw angle of camera 26. On the left / right scale, the numerical value corresponding to reference line 802 represents the pitch angle of camera 26. The pitch angle indicates 90° when camera 26 is pointed vertically downward.

[0077] In one implementation example, the application program 221 sets a downward 45° angle as the appropriate pitch angle. The appropriate pitch angle can also be set to 30°, 60°, or 90°. Corresponding to this setting, in the example of FIG. 8, the number "45" is emphasized relative to the other numbers ("20," "30," "40," "50," "60," and "70") on the left and right scales, thereby emphasizing the downward 45° angle. In this specification, the pitch angle of the camera 26 refers to the angle between the optical axis of the camera 26 and the horizontal direction. The pitch angle of the camera 26 may also be simply referred to as the angle of the camera 26.

[0078] Returning to FIG. 7, when the user 5 presses the capture button, the terminal 2 captures an image in step S32.

[0079] In step S34, the terminal 2 transmits to the server 1, as prediction information, the captured image, and the date, time, and place (location information) at which the image was captured.

[0080] In response to receiving the calculation information from the terminal 2, the server 1 starts the process of FIG.

[0081] In step S40, the server 1 receives the prediction information transmitted from the terminal 2.

[0082] In step S42, the server 1 selects a prediction model.

[0083] More specifically, a plurality of prediction models corresponding to a plurality of varieties of paddy rice may be stored as the prediction model 123. The prediction information may include information indicating the variety specified by the user 5. In step S42, the server 1 may select, from the plurality of prediction models, a prediction model corresponding to the variety specified by the user 5. If a single prediction model is stored in the server 1, the prediction model is identified in step S44.

[0084] In step S44, the server 1 converts the image included in the prediction information into a vegetation cover ratio.

[0085] In step S46, the server 1 applies (inputs) the vegetation coverage rate obtained in step S44 to the prediction model selected (or specified) in step S42.

[0086] In step S48, the server 1 acquires, from the prediction model, the prediction result of the remaining accumulated temperature corresponding to the vegetation coverage ratio obtained in step S44.

[0087] In step S50, the server 1 acquires, from the weather information server 3, weather information corresponding to the location in the prediction information.

[0088] In step S52, the server 1 identifies a suitable day for mid-drying based on the date and time when the image was taken, the predicted result of the remaining accumulated temperature acquired in step S48, and the weather forecast acquired in step S50. More specifically, the server 1 identifies the day when the accumulated temperature from the day when the image was taken reaches the predicted result of the remaining accumulated temperature as a suitable day for mid-drying.

[0089] In step S54, the server 1 transmits the suitable day for mid-drying identified in step S52 to the terminal 2 as a prediction result, and ends the processing of FIG.

[0090] Meanwhile, the terminal 2 receives the prediction result transmitted from the server 1 in step S36.

[0091] Then, in step S38, the terminal 2 displays the prediction result and ends the processing of FIG.

[0092] In the process described above, the user 5 uses the terminal 2 to take an image of the paddy rice being grown in the field and transmits the image as prediction information to the server 1. As a result, the user 5 is provided with the prediction result of the suitable day for mid-season drying of the paddy rice in the field from the server 1 as a display on the terminal 2.

[0093] [Rice cultivation method using the method for predicting the optimal day for mid-season drying according to the present disclosure] In the prediction system of the present disclosure, the user 5 can cause the server 1 to execute a method for predicting suitable days for mid-season drying, and use the prediction results of suitable days for mid-season drying derived according to the prediction method for rice cultivation.

[0094] More specifically, on a given day after planting, the user 5 starts the application program 221 on the terminal 2 and performs an operation to predict the suitable day for mid-season drying. At this time, the user 5 may input the rice variety and planting date into the application program 221.

[0095] In response to the above operation, the terminal 2 instructs the camera 26 to photograph the field. The user 5 photographs the field. The photograph of the field may be taken from the periphery of the field or by entering the field.

[0096] When photographing the farm field, the user 5 may refer to the scale attached to the viewfinder image as described with reference to Fig. 8. More specifically, the user 5 may press the capture button when he or she determines, by referring to the scale, that the pitch angle of the camera 26 at that time is 45° downward. This pitch angle may be 30°, 60°, 90° downward, or the like.

[0097] When the capture button is pressed, the terminal 2 sends "prediction information" including the captured image to the server 1, as described with reference to Figure 7, receives the predicted results of the suitable days for mid-term drying from the server 1, and displays the predicted results of the suitable days for mid-term drying.

[0098] Before transmitting the "prediction information" to the server 1, the terminal 2 may have the user 5 confirm the image to be transmitted as the "prediction information." More specifically, when the capture button is pressed, the terminal 2 may display the image captured at that time on the display 24 along with two menus (transmission possible, retake capture). In response to the user 5 selecting the menu "transmission possible," the terminal 2 may transmit the "prediction information" including the captured image to the server 1. In response to the user 5 selecting the menu "retake capture," the terminal 2 may display a viewfinder image on the display 24 to allow the user 5 to capture an image of the field again.

[0099] The user 5 recognizes the predicted results of the suitable day for mid-drying from the display on the terminal 2. The user 5 then carries out mid-drying at the timing specified based on the predicted results, and also carries out the subsequent processing for cultivating paddy rice. The user 5 may record information about the cultivation record (the date when mid-drying started, the date when mid-drying ended, images of the field, etc.) on the terminal 2.

[0100] [Analysis of the correlation between vegetation cover and residual accumulated temperature] In this embodiment, the optimum date for mid-season drying of rice plants in a field is predicted using images taken in the field. Four types of processing patterns were defined for the images used for the prediction, and the correlation between vegetation cover and residual accumulated temperature for each processing pattern was compared.

[0101] Each image was a still image of 4032 x 3024 pixels before processing. The entire image to be converted into vegetation cover was converted from RGB to CIE lab color space and binarized.

[0102] 9 is a diagram showing images after processing using four types of processing patterns, together with the image before processing. In FIG. 9, image 90 represents the image before processing, and images 91 to 94 represent images according to the four types of processing patterns. Each of images 91 to 94 is a CIEa * The following shows the processed images when four thresholds (-5, -10, -15, Otsu) are set. Otsu's threshold is based on a method of creating a histogram of the brightness of each pixel in the image, with the pixel value on the horizontal axis and the number of pixel values ​​on the vertical axis, and then dividing the image into two groups at a certain pixel value to set the threshold at the pixel value that maximizes the L-distance (inter-class variance / intra-class variance).

[0103] Fig. 10 shows images after processing using four different processing patterns. Images 94 to 97 in Fig. 10 differ from one another in the content of cropping processing performed on the images binarized according to Otsu's threshold.

[0104] No cropping has been performed on image 94 in Fig. 10. No cropping has been performed on images 91 to 94 in Fig. 9 either. That is, it is intended that image 94 in Fig. 10 and image 94 in Fig. 9 have the same processing conditions.

[0105] In image 95 of FIG. 10, some pixels in the long axis direction have been deleted by cropping. More specifically, in image 95 of FIG. 10, one-quarter (1008 pixels) of the pixels at the top (the farther side when the user who captured the image is used as the reference) have been deleted by cropping. In image 96 of FIG. 10, one-quarter (1008 pixels) of the pixels at the bottom (the nearer side when the user who captured the image is used as the reference) have been deleted by cropping. In image 97 of FIG. 10, one-quarter (1008 pixels) of the pixels at both the top and bottom have been deleted by cropping.

[0106] In the following description, the seven types of processing patterns corresponding to the images 91 to 97, respectively, may be referred to as Pat.1 to 7. For example, the processing pattern corresponding to the image 91 is referred to as Pat.1.

[0107] The following description also refers to the state in which a user takes an image.

[0108] 11 is a diagram for explaining three types of states in which an image is captured. In FIG. 11, the three types of states in which an image is captured are shown within frames 1101 to 1103, respectively.

[0109] Frame 1101 shows a state in which a user is photographing a field from the edge of the field, with the camera's pitch angle pointing downward at 45 degrees. Frames 1102 and 1103 show states in which a user is entering the field and taking images. In the state of frame 1102, the camera's pitch angle is pointing downward at 45 degrees, and in the state of frame 1103, the camera's pitch angle is pointing downward at 90 degrees.

[0110] For each of the seven processing patterns described above, the coefficient of determination R is calculated when the images taken in the conditions of frames 1101, 1102, and 1103 are used as the images converted to vegetation cover ratios. 2 can be calculated.

[0111] The coefficient of determination R 2 To calculate this, we first create a scatter plot of the vegetation coverage converted from the image and the residual accumulated temperature, which is the predicted result, and then perform linear regression using the least squares method. Then, we calculate the correlation coefficient r of the linear regression according to the following formula (1), and calculate the coefficient of determination R as the square of the correlation coefficient r. 2 can be calculated.

[0112]

number

[0113] In equation (1), the explanatory variable x represents the natural logarithm of the vegetation coverage rate, and the dependent variable y represents the residual accumulated temperature. The subscript i represents the value that identifies each image. The explanatory variable x and dependent variable y with a "-" above them represent the average value of the explanatory variable x and dependent variable y, respectively.

[0114] The coefficient of determination R2 The model (prediction model 123) that predicts the residual accumulated temperature from the vegetation cover ratio used to calculate the residual accumulated temperature is a model that predicts the residual accumulated temperature by linear regression from the natural logarithm of the vegetation cover ratio.

[0115] [Comparison of prediction errors for the best days for mid-drying] It is possible to compare prediction errors for the optimum days for mid-drying predicted by the system according to this embodiment.

[0116] The system can use a model similar to that mentioned in the above [Analysis of the correlation between vegetation coverage rate and residual accumulated temperature] as a prediction model for residual accumulated temperature to predict suitable days for mid-drying (a model that predicts residual accumulated temperature by linear regression using the natural logarithm of vegetation coverage rate (coverage) extracted from the image [logarithmic vegetation coverage rate (log_coverage)]).

[0117] By comparing the residual accumulated temperature predicted by the prediction model with the average daily temperature data for the past five years, the day on which the accumulated temperature from the shooting date is closest to the predicted value of the residual accumulated temperature can be identified as the predicted day for mid-drying. Note that 1km mesh weather data or weather forecast data can be used as temperature data. Of the weather data or weather forecast data, only the temperature information can be used.

[0118] The number of days obtained by subtracting the predicted suitable day for mid-season drying from the true suitable day for mid-season drying can be determined as the prediction error (days).

[0119] The prediction error used for evaluation can be RMSE (Root Mean Square Error) calculated according to the following equation (2).

[0120]

number

[0121] In equation (2), y i represents the i-th predicted result.i The symbol "^" above the symbol indicates the true optimum day for mid-season drying corresponding to the i-th forecast result. N indicates the number of forecast results.

[0122] A dataset generated from images taken in multiple fields can be used to train the predictive model used in the system being evaluated. If the dataset prepared for predicting the optimum days for mid-season drainage for system evaluation is the same as the dataset prepared for generating the predictive model, the RMSE can be calculated by leave-one-field-out cross validation (LOFO-CV) and is denoted as "RMSEcv."

[0123] In LOFO-CV, one field's worth of data set is used as test data, and the remaining fields' worth are used as training data. LOFO-CV will be described in more detail with reference to FIG. 12.

[0124] Fig. 12 is a diagram for explaining leave-one-field cross-validation. In the example of Fig. 12, the dataset includes data for three fields indicated by "A," "B," and "C."

[0125] The example in FIG. 12 includes three verification steps (first verification step to third verification step).

[0126] More specifically, in the first verification step, the data indicated by "A" and "B" are used as training data to generate a prediction model, and then the data indicated by "C" is used as test data to calculate the RMSE for the optimal day for mid-drying.

[0127] Next, in the second verification step, the data indicated by "B" and "C" are used as training data to generate a prediction model, and the data indicated by "A" is used as test data to calculate the RMSE for the optimal day for mid-drying.

[0128] Next, in the third verification step, the data indicated by "C" and "A" are used as training data to generate a prediction model, and the data indicated by "B" is used as test data to calculate the RMSE for the optimal day for mid-drying.

[0129] After the three RMSEs are calculated as described above, the average of the three RMSEs is calculated. In LOFO-CV, the average calculated in this way is identified as RMSEcv.

[0130] In addition, if the number of stems used as a guideline for mid-drying differs between the training data and the test data, the number of stems used as a guideline for mid-drying in the test data is applied to the training data, the remaining accumulated temperature is recalculated, and a prediction model is generated, and the prediction model generated in this manner can be used to predict the remaining accumulated temperature.

[0131] In the present invention, any of early-maturing, mid-maturing, and late-maturing varieties of paddy rice can be used, such as ICS6, Koshihikari Tsukuba SD1, Tsukuba SD2, Koshihikari, Aikoku, Aichi no Kaori, Akage, Akanesora, Akisakari, Akitakomachi, Akidawara, Akihikari, Akebono, Asa no Hikari, Asahi, Asahi, Asahi no Yume, Itadaki, Eiko, Eiko, Egami no Kizuna, Oidemai, Ouu 197, Oba, Okiniiri, Otomomechi, Oborozuki, Omachi, Shinriki, Kameji, Kame no O, Kame no O 4, Kitaaobaba, Kinuhikari, Kinumusume, Kiho, Kirara 397, Ginbozu, Kinmikaze, and Kusanoho. Shi, Premonition of Love, Golden Clear, Koganemasari, Koganemochi, Koshiji Wase, Koshihikari, Gohyakumangoku, Goropikari, Sasashigure, Sasanishiki, Satojiman, Tokoku, Joshu, Shirosenbon, Sekitori, Senichi, Wind of the Earth, Star of the Earth, Taichung 65, Takanari, Takenari, Tachiaoba, Tachiayaka, Tachijobu, Tachisuzuka, Haruka, Chiyonishiki, Chiyohonami, Tsugaru Roman, Moonlight, Tsuyahime, Todorokiwase, Domannaka, Toyonishiki, Dontokoi, Chusei Shinsenbon, Natsuhikari, Nanatsuboshi, Nankoku So Dachi, Nikomaru, Nishihomare, Nipponbare, Nihonmasari, Norin 18, Norin 1, Norin 22, Norin 29, Norin 6, Norin 8, Notohikari, Haigokoro, Haenuki, Hatsushizuku, Hatsushimo, Hatsunishiki, Hatsuhoshi, Hanaechizen, Hayamasari, Haruru, Hitomebore, Hinohikari, Himegomi, Himenomochi, Hyakumankoku, Hiyokumochi, Fukuhibiki, Fukei 175, Fusaotome, Fujisaka 5, Fujihikari, Futaba, Fukkuriinko, Benisengoku, Bozu, Honenwase, Hoyoku, Some of the varieties include Hokuriku 193, Hoshiaoba, Hoshino Yume, Hoshimaru, Hohohonoho, Matsuribare, Manamusume, Mizuho Chikara, Mizuho no Kagayaki, Mineasahi, Mineharuka, Mirenishiki, Mutsuhomare, Menkoina, Moeminori, Momiroman, Morita Wase, Yamasenishiki, Yamada Nishiki, Yamadawara, Yamabiko, Yukar, Yuki no Sei, Yukihikari, Yukimaru, Yumeakari, Yumetsukushi, Yumehikari, Yume Hitachi, Yumepirika, Yumemizuho, ​​Rikuu 132, Rikuu 20, Reihou and Reimei. [Explanation of symbols]

[0132] 1 Server, 2 Terminal, 3 Weather information server, 5 User, 11,21 Controller, 12,22 Memory, 13,23 Communication interface, 14,24 Display, 15,25 Input device, 26 Camera, 27 GPS receiver, 28 Acceleration sensor, 120,220 Data, 121,221 Application program, 122 First program, 123 Prediction model, 124 Second program, 151 Acquisition unit, 152 Conversion unit, 153,154 Prediction unit, 1101,1102,1103 Frame, IM Image, N Network.

Claims

1. A method for predicting a suitable day for mid-season drying of paddy rice in a field, comprising: A computer acquires an image of the field; The computer converts the image into a vegetation cover in the field; and a step of the computer predicting a residual accumulated temperature in the field based on the vegetation cover ratio, The remaining accumulated temperature represents the accumulated temperature required until the suitable day for mid-drying, A prediction method comprising a step in which the computer generates a prediction result of the suitable day for mid-drying based on the period from the time the image was acquired until the accumulated temperature reaches the remaining accumulated temperature by referring to weather information corresponding to the field.

2. predicting the residual accumulated temperature includes applying the image to a prediction model; The prediction model is configured to output a prediction result of the residual accumulated temperature by inputting a vegetation cover ratio through machine learning processing using training data, The prediction method of claim 1 , wherein each of the two or more data sets included in the training data includes vegetation coverage tagged with residual accumulated temperature.

3. a step of acquiring by the computer the variety of paddy rice in the field; The prediction method according to claim 2 , further comprising a step of selecting, by the computer, as the prediction model, a model corresponding to the variety from two or more models.

4. The computer acquires location information of the field; The prediction method according to claim 1 , further comprising the step of acquiring weather information corresponding to the location information as the weather information corresponding to the field.

5. A step in which a terminal captures the image; the terminal transmitting the image to the computer; The prediction method according to claim 1 , wherein in the step of capturing an image, information relating to an angle at which the terminal captures the image is output in the terminal.

6. A method for cultivating paddy rice, comprising utilizing the prediction method according to any one of claims 1 to 5.

7. A prediction program for predicting a suitable day for mid-season drying of paddy rice in a field, The prediction program is executed by a processor of a computer, thereby causing the computer to: acquiring an image of the field; converting the image into a percent vegetation cover in the field; and predicting a residual accumulated temperature in the field based on the vegetation coverage ratio; The remaining accumulated temperature represents the accumulated temperature required until the suitable day for mid-drying, The prediction program causes the computer to perform a step of generating a prediction result of the suitable day for mid-drying based on the period from the time the image was acquired until the accumulated temperature reaches the remaining accumulated temperature by referring to weather information corresponding to the field.

8. A system for predicting a suitable day for mid-season drying of paddy rice in a field, comprising a terminal and a computer, The terminal a camera that captures an image of the field; a first communication interface for transmitting the image to the computer; The computer a second communication interface for acquiring the image; a controller that converts the image into a vegetation coverage ratio in the field; The controller Predicting a residual accumulated temperature in the field based on the vegetation coverage rate; The residual accumulated temperature is represents the accumulated temperature required until the suitable day for mid-drying, The controller By referring to meteorological information corresponding to the field, a prediction result of the suitable day for mid-season drying is generated based on the period from the time when the image was acquired until the accumulated temperature reaches the remaining accumulated temperature; The system transmits the prediction result to the terminal.

9. A method for generating a model for predicting the remaining accumulated temperature, which is the temperature required until the suitable date for mid-drying of paddy rice in a field, acquiring one or more images of the field; each of the one or more images is associated with an accumulated temperature of the field; Identifying a suitable day for mid-season drainage in the field where the one or more images were taken; A step of generating a data set by tagging the vegetation coverage rate of each of the one or more images as the remaining accumulated temperature, which is a value obtained by subtracting the accumulated temperature of each of the one or more images from the total accumulated temperature, which is the accumulated temperature corresponding to a suitable day for mid-season drying in the field; A method for generating a model comprising: using the data set of each of the one or more images as training data to perform machine learning on a model that outputs residual accumulated temperature when vegetation coverage is input.

10. The model generating method according to claim 9 , wherein in the step of identifying the suitable day for mid-drying, the total accumulated temperature is identified based on the number of stems.

Images