JP2017169511A - Apparatus for estimating normal stock ratio of agricultural crop, apparatus for predicting yield of agricultural crop, and method for estimating normal stock ratio of agricultural crop - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide apparatus for estimating normal stock ratio, which early predicts reduction in harvest amount due to missing stock and predicts the yield of crops with high accuracy.SOLUTION: According to the present invention, an apparatus comprises: a prediction target-cropping setting section 21-1 for setting a cropping ID of a prediction target: an image data-analyzing section 21-2 for calculating a vegetation index representative value of a field of prediction target-cropping on an image photographing date; a growth model-adapting section 21-3 for calculating a representative value of the estimated vegetation index; a growing stage-estimating section 21-4 for estimating a growing stage at every elapse of days from a cropping start date, an image data-selecting section 21-6 for selecting image data of a second resolution; an image data-analyzing section 21-7 for calculating a multidimensional vector for each pixel of the field of prediction target cropping in the image of the second resolution; and a normal stock ratio-estimating section 21-8 for specifying, from clusters obtained by classifying multidimensional vectors, a cluster in which the pixels of the normal stock portion are classified, and estimating a normal stock ratio in the field from the ratio of the pixels belonging to the specified cluster to all pixels in the field.SELECTED DRAWING: Figure 1B

Description

  The present invention relates to a crop yield prediction (normal stock rate estimation) technique for crops.

  Agricultural crop supply and prices can fluctuate significantly due to weather and other factors. For example, in order to procure stable farm products, a consumer makes a contract transaction in which a contract is made with a producer in advance regarding the shipping amount, price, shipping date, and the like.

  However, if the production volume of the entire contract producer is less than the planned procurement amount, the consumer must procure crops separately to reach the planned quantity, for example, by importing crops from abroad. Accordingly, the consumer needs to accurately predict the crop yield of the contract producer and appropriately determine the quantity of agricultural products to be procured separately.

  By the way, in the case of crops where the harvested part grows in the ground, such as root vegetables, conduct a field survey before harvesting, and investigate the rate of stock loss in the field and the weight of the harvested part per share. So we were predicting the yield.

  However, yield prediction by this method is time consuming and inefficient when there are many fields of contract producers and they are distributed over a wide area. In addition, since the field survey is performed only in a part of the section of the field, there is a problem in that it cannot be accurately predicted if there is a variation in the stock loss rate or the weight per strain in the field.

  In order to solve such problems, a method for predicting the crop yield from satellite images observed by remote sensing using artificial satellites has been considered.

  For example, in Non-Patent Document 1 below, a vegetation index such as NDVI (Normalized Difference Vegetation Index) is calculated from a satellite image of a prediction target area observed by remote sensing using an artificial satellite, and the value of NDVI accumulated over a certain period is used. The crop yield in the area is predicted.

  Further, in the vegetation growth analysis system described in Patent Document 1 (for example, claim 6), vegetation indices such as NDVI and vegetation growth curves are corrected using low-resolution satellite data and a small number of high-resolution satellite data. . Thereby, the accuracy of yield prediction based on the vegetation index can be improved.

  In Non-Patent Document 2 below, using a field survey result as teacher data, a classification tree for estimating the degree of disease of crops is learned from resolution satellite images, and the estimated disease occurrence status in the field and the actual yield Analyzing the relationship.

  In the vegetation growth analysis system described in Patent Document 1 (for example, claim 2), each point cloud of the high-dimensional analysis model data is obtained by using past disaster data stored in advance as teacher data with respect to disasters such as lodging and diseases. To learn the closed surface / line that separates data with disaster from normal growth data. Thereby, it can be determined from the satellite image whether or not the growth of the target plant is normal, and the result can be applied to yield prediction.

Japanese Patent Laying-Open No. 2015-188333

CJ Weissteiner, W. Kuhbauch, Regional Yield Forecasts of Malting Barley (Hordeum vulgare L.) by NOAA‐AVHRR Remote Sensing Data and Ancillary Data, Journal of Agronomy and Crop Science, Vol. 191 Issue 4, pp.308-320 (2005 ). Jonas Franke, Gunter Menz, Multi-temporal wheat disease detection by multi-spectral remote sensing, Precision Agriculture, Vol. 8, Issue 3, pp. 161-172 (2007).

  However, in the techniques described in Non-Patent Document 1 and Patent Document 1, the crop yield is predicted based only on the vegetation information on the ground part. For this reason, it is difficult to predict a decrease in yield due to a lack of stock at an early stage, and there is a problem that a prediction error becomes large particularly in a crop where the harvested portion grows in the ground.

  Further, in the methods described in Non-Patent Document 2 and Patent Document 1, teacher data is required for learning a classification model. In these documents, in the case of estimation of a missing stock, a classification model for discriminating whether the satellite image data is a normal stock or a missing stock is obtained by using satellite image data labeled normal stock or stock missing stock as teacher data. learn.

  Here, the vegetation of the above-ground part of the crop varies depending on the stage of growth. Therefore, even if the same normal strain is used, the observed satellite image data differs depending on the stage of growth.

  Therefore, it is necessary to prepare different teacher data for each stage of growth and learn the classification model. However, it is difficult to obtain a sufficient amount of teacher data for this from actual field surveys and satellite image analysis. There is.

  An object of the present invention is to provide a technique capable of predicting a decrease in yield due to a lack of stock at an early stage and predicting a crop yield with high accuracy.

  According to one aspect of the present invention, there is provided a normal stock rate estimation device for estimating a normal crop rate of a crop, a cropping ID for uniquely identifying crop cropping, and a field for uniquely identifying a cropped field The cropping information in which the ID, the cropping start date that is the date when cropping is started, and the cropping end date that is the date that crops are harvested and cropping is to be completed, Registered are GIS information for each field in which information representing the position and shape is registered, a growth stage representing the cropping and growing stages of the crop, and information representing whether or not the image data analysis target growth stage. Growth stage information, an image ID for uniquely identifying an image of the first resolution, an image shooting date that is a date when the image of the first resolution is shot, and an image data body of the first resolution And is registered Image information of the first resolution, an image ID that uniquely identifies the image of the second resolution, an image shooting date that is a date when the image of the second resolution is shot, and image data of the second resolution A main body and a storage device that stores image information of the second resolution in which the main image is registered, a prediction target cropping setting unit that sets a crop target ID of the prediction target from the input device, and the set Based on the cropping ID, referring to the cropping information and the image information of the first resolution, from the image data of the first resolution, the representative vegetation index value of the field to be predicted cropped on the image shooting date An image data analysis unit that calculates a growth model adaptation unit that refers to the growth stage information and calculates an estimated vegetation index representative value by adapting the growth model of the crop to the time series data of the vegetation index representative value; A growth stage estimation unit that estimates a growth stage for each elapsed day from the start date of cropping based on at least one of the change rate and the number of days of the adapted crop model, and analysis for each growth stage of the image data analysis target The image data selection unit for the second resolution that selects the image data of the second resolution to be the target, and the second pixel that is the analysis target for each pixel of the field to be predicted cropped in the image of the second resolution An image data analysis unit for a second resolution that calculates a multidimensional vector by combining pixel-specific vegetation indexes calculated from image data of a resolution of the above, and classifying the multidimensional vector by clustering from only the vector value. Identify one or more clusters in which the pixels of the normal strain are classified according to the specified conditions There is provided a normal stock rate estimation device comprising: a normal stock rate estimation unit that estimates a normal stock rate in a field from a ratio of pixels belonging to the specified cluster to all pixels.

  According to the normal stock rate estimation apparatus, the multidimensional vector is classified by clustering from only the value of the vector, and unsupervised learning is performed in order to identify a cluster in which the pixels of the normal stock portion are classified according to a predetermined condition. can do.

The first resolution is preferably lower than the second resolution. The normal stock rate estimation unit may specify a cluster into which pixels of the normal stock portion are classified based on a condition with respect to the center point of the cluster. As an example, a method of specifying a cluster of normal strain pixels can be used by selecting _ck having the largest vegetation index value in the growth stage of “vegetation stop”.

  In the normal strain, the vegetation index increases until the growth stage of “vegetation stop”, whereas in the lacking part, the vegetation index does not increase from the middle of the growth stage of “vegetation increase”. In addition, the vegetation index of the soil portion where the strain has not been planted continues from the growth stage “immediately after planting”.

  The vegetation index representative value and the vegetation index by pixel are each NDVI, EVI, or GRVI.

  The image data analysis unit extracts an image of the field from the image data of the second resolution, performs coordinate conversion so that the position and orientation of the field match, and outputs the image of the field after coordinate conversion processing. For each pixel P, it is preferable to calculate a vegetation index VI (S, P) at the pixel P.

  The present invention is a yield predicting device for predicting the yield of agricultural products, the normal stock rate estimating device described above, and a yield predicting model including at least the normal stock rate as an explanatory variable. A yield predicting device having a yield predicting unit for predicting the yield may be used. Here, the estimated rate may be a stock loss rate.

  In the growth stage information, information indicating whether or not the growth stage is a growth stage subject to growth aggregation is registered, and further, the estimated vegetation index representative value of the prediction target crop in the growth stage subject to growth aggregation is summed And calculating the estimated growth degree of the target crop in step, comprising a growth degree estimation unit, the yield prediction unit based on at least the estimated growth degree and the yield prediction model including the normal stock rate as an explanatory variable, It is characterized by predicting the yield of the target crop. Here, the estimated vegetation index representative value is, for example, a median value, a mode value, or an average value of the vegetation index.

  According to another aspect of the present invention, there is provided a normal stock rate estimation method for estimating a normal crop rate of a crop, and a crop ID that uniquely identifies crop cropping and a field on which cropping has been performed are uniquely identified. For each field, the field ID, the crop start date that is the date when the crop is started, and the crop end date that is the date that the crop is harvested and the crop is scheduled to end, GIS information by field, information indicating the cropping and growth stages of crops, and information indicating whether or not the image data analysis target growth stage is registered. Registered growth stage information, an image ID for uniquely identifying an image of the first resolution, an image shooting date that is a date when the image of the first resolution is shot, and image data of the first resolution The main unit is registered First resolution image information, an image ID that uniquely identifies the second resolution image, an image shooting date that is the date when the second resolution image was shot, and the second resolution image. A prediction target cropping setting step for setting the cropping ID of the prediction target from the input device with reference to each piece of information of the storage device that stores the image data of the second resolution in which the data body is registered Based on the cropping ID, the vegetation index of the field to be predicted cropped on the date of image capture from the image data of the first resolution acquired with reference to the cropping information and the image information of the first resolution An image data analysis step for calculating a representative value and a growth model that calculates an estimated vegetation index representative value by referring to the growth stage information and adapting the growth model of the crop to the time series data of the vegetation index representative value. A adapting step, a growing stage estimating step for estimating a growing stage for each elapsed day from the start date of cropping based on at least one of the change rate and the number of days of the growing model of the adapted crop, and a growing stage for image data analysis For each second image data selection step for selecting the second resolution image data to be analyzed, and for each pixel in the field of the prediction target cropping in the second resolution image, An image data analysis step for a second resolution for calculating a multidimensional vector by combining pixel-specific vegetation indices calculated from the image data of the second resolution, and clustering the multidimensional vector from only the value of the vector Classify the cluster where the pixels of the normal stock are classified from the obtained cluster. A normal strain ratio, wherein at least one normal strain is identified by a predetermined condition, and a normal stock rate estimation step for estimating a normal stock rate in the field from a ratio of pixels belonging to the identified cluster to all pixels of the field. A rate estimation method is provided.

  The present invention may be a program for causing a computer to execute the normal stock rate estimation method described above, or a computer-readable storage medium for recording the program.

  ADVANTAGE OF THE INVENTION According to this invention, the technique which can estimate the reduction | decrease in the yield by a lacking strain at an early stage, and can estimate the crop yield of a crop with high precision can be provided.

It is a functional block diagram which shows the schematic structural example of the harvest amount prediction apparatus by embodiment of this invention. FIG. 1B is a functional block diagram illustrating a more detailed configuration example of the configuration of FIG. 1A. It is a figure which shows one structural example of a cropping information table. It is a figure which shows the example of 1 structure of the GIS information table classified by field. It is a figure which shows the example of 1 structure of a growth stage table. It is a figure which shows one structural example of a low-resolution image information table. It is a figure which shows the example of 1 structure of a high resolution image information table. It is a figure which shows the example of 1 structure of the vegetation index table classified by planting. It is a figure which shows the example of 1 structure of an estimated growth degree table. It is a figure which shows one structural example of the analysis result table classified by pixel. It is a figure which shows the example of 1 structure of an estimated normal stock ratio table. It is a figure which shows the example of 1 structure of a predicted harvest amount table. It is a figure which shows the example of 1 structure of a performance harvest amount table. It is a flowchart figure which shows the flow of the crop yield prediction process of the crop by this Embodiment. It is a flowchart figure which shows the flow of the resolution image data analysis process in a low resolution image data analysis part. It is a flowchart figure which shows the flow of the growth model adaptation process in the growth model adaptation part. It is a flowchart figure which shows the flow of the growth degree estimation process in a growth degree estimation part. It is a flowchart figure which shows the flow of a process in a high resolution image data selection part. It is a figure following FIG. 17A. It is a flowchart figure which shows the flow of the high resolution image data analysis process in a high resolution image data analysis part. It is a flowchart figure which shows the flow of the normal stock rate estimation process in a normal stock rate estimation part. It is a figure which shows the image of the growth model adaptation process in the growth model adaptation part. It is a figure which shows the image of the growth stage estimation process in a growth stage estimation part. It is a figure which shows the image of the growth degree estimation process in a growth degree estimation part. It is a figure which shows the image of the high resolution image data selection process in a high resolution image data selection part. It is a figure which shows the image of the elapsed days selection process of a high resolution image data selection part. It is a figure which shows the image of a coordinate transformation process. It is an image figure of the process which calculates the vegetation index VI (S, P) classified by pixel in the pixel P. It is a figure which shows the image of the 1st cluster processing example. It is a figure which shows the image of the 2nd cluster processing example.

  Hereinafter, a yield prediction technique based on a normal crop rate estimation technique for a crop according to an embodiment of the present invention will be described in detail with reference to the drawings, taking root vegetables as an example of the crop. However, the crops are not limited to root vegetables.

  First, a system configuration example of a crop yield prediction device according to the present embodiment will be described. FIG. 1A is a functional block diagram illustrating a schematic configuration example of a yield prediction apparatus A according to the present embodiment. As shown in FIG. 1A, the yield prediction apparatus A includes a database group 1 having databases 1a, 1b, 1c,... For storing information used for yield prediction, and a prediction calculation for performing a yield prediction calculation. Part B and a network NT such as the Internet for associating them are included. The prediction calculation unit B includes various interface units C, a control unit 7, and a memory D that stores a program for performing calculation processing. The yield prediction device A and the prediction calculation unit B may be composed of, for example, a personal computer PC, and may have an integrated configuration without using the Internet.

  FIG. 1B is a functional block diagram illustrating a more detailed configuration example of the configuration example of FIG. 1A. As shown in FIG. 1B, the yield prediction device A according to the present embodiment includes an auxiliary storage device 1 corresponding to the database group 1 and various interface units C, for example, an input device 3 such as a keyboard, a display, and the like. Output device 5 and a central processing unit (CPU) 7, and further includes a main storage device 21 for recording a program for realizing various processing units by software configuration. Various processing units may be configured by a hardware configuration.

  More specifically, the main storage device 21 and the CPU 7 can realize the following processes in various processing units.

  The weather information acquisition unit 11 acquires a weather forecast from an external system via the Internet or the like. The weather forecast includes long-term forecasts such as a one-month forecast and a two-month forecast. The weather information acquisition unit 11 also acquires past weather observation results (actual data). Details thereof will be described later.

≪Various information≫
Various types of information are stored in the memory in a table format as described below, for example. 2 to 12 are diagrams showing examples of data structures of various tables.

<Crop information table 11-1>
As illustrated in FIG. 2, the crop information table 11-1 is a table in which information on crops that are targets for at least yield prediction is registered.

For example, the following data is stored.
The cropping ID that uniquely identifies the crop cropping, the field ID that uniquely identifies the cropping field, the cropping start date that is the date the cropping was started, and the date that the crop is to be harvested and the cropping is scheduled to end The planned end date of planting, the number of planted stocks that are the number of plants planted in the planting.

<Field-specific GIS information table 11-2>
As shown in FIG. 3, the field-specific GIS information table 11-2 is a table in which information indicating the position and shape on the map of the field where cropping is being performed is registered.

For example, the following data is stored.
A field ID, a shape file storing information representing the position and shape of the field on the map, and the area of the field. Here, the shape file is a data format in which the shape of the field can be expressed by polygons, and the position and shape of the field on the map can be expressed by using longitude and latitude as the coordinate system. The area of the field can be used when it is desired to obtain the yield per unit area. If necessary, the area may be obtained from the shape file and registered in advance.

<Growth stage table 11-3>
As shown in FIG. 4, the growth stage table 11-3 is a table in which each growth stage of the crop and information accompanying it are registered.

For example, the following data is stored.
Estimated vegetation index when estimating the degree of growth in the growth stage such as “immediately after planting”, “vegetation increase”, “vegetation stop”, “vegetation decrease”, etc. Information about whether or not it is a growth level target indicating whether or not it is a growth stage that is the target of aggregation of average values, whether or not it is a target growth stage when selecting a high-resolution image in the high-resolution image data selection section described later The same information based on the number of days that have elapsed since the start date of cropping when selecting a high-resolution image in the high-resolution image data selection unit to be described later. Elapsed days selection formula, etc. for selecting an image to be analyzed from a plurality of high resolution images in the growth stage.

  It should be noted that the elapsed day selection formula need not be registered in a record in which “No” is registered in the column for high resolution image data analysis.

As an example of the elapsed days selection formula, the following formula can be used. Here, S is the growth stage, D _START (S) is the first elapsed day of the growth stage S, D _END (S) is the last elapsed day of the growth stage S, {D _n (S)} _{n = 1,. , N} is the number of days of the growth stage S at which a high-resolution image can be acquired, and satisfies D _START (S) ≦ D _n (S) ≦ D _END (S).

  Here, Expression (1) is an expression for selecting the earliest time from the elapsed days of the growth stage S where a high-resolution image can be acquired.

Equation (2) is the closest to (D _END (S)-D _START (S)) / 2, which is exactly half the elapsed days in the growth stage S from the elapsed days in the growth stage S where high-resolution images can be acquired. It is an expression that selects things. Expression (3) is an expression for selecting the latest one from the elapsed days of the growth stage S where a high-resolution image can be acquired.

<Low Resolution (First Resolution) Image Information Table 11-4>
As shown in FIG. 5, the low resolution image information table 11-4 is a table in which information on low resolution satellite images used for yield prediction is registered.

For example, the following data is stored.
A low resolution image ID that uniquely identifies the low resolution image, an image shooting date that is the date when the low resolution image was shot, and an image file that is a file of the low resolution image data body.

  The satellite image is obtained by a multispectral observation satellite sensor including a band of visible light and near-infrared light that can detect crops, particularly leaves appearing on the ground surface, such as Landsat, and the field is identified. Use what was acquired at a resolution that is possible. However, the type of satellite image does not have to be limited to a low resolution actually.

<High Resolution (Second Resolution) Image Information Table 11-5>
As shown in FIG. 6, the high resolution image information table 11-5 is a table in which information on high resolution satellite images used for yield prediction is registered.

For example, the following data is stored.
A high resolution image ID that uniquely identifies a high resolution image, an image shooting date that is the date when the high resolution image was shot, and an image file that is a file of the high resolution image data body.

  Satellite images were acquired with multispectral observation satellite sensors including visible and near-infrared bands, such as GeoEye, and within one pixel so that the status of normal and missing strains in the field can be estimated. Use as few strains as possible. However, the types of satellite images are not limited in the same manner.

  Here, the distinction between the low resolution and the high resolution is based on a difference in relative resolution, and the available satellite images are not limited by the absolute resolution. Furthermore, an image acquired by the same observation satellite sensor as the high-resolution satellite image may be used as the low-resolution satellite image.

<Vegetation index table 11-6 by cropping>
As shown in FIG. 7, the vegetation index table 11-6 according to cropping is a table for registering information on the vegetation index average value of the field for each cropping and for each elapsed day from the cropping start date.

For example, the following data is stored.
Planting ID, number of days elapsed from the planting start date of the planting, average vegetation index, estimated vegetation index average of the field in the elapsed days calculated by the low-resolution image data analysis unit described later, growth stage estimation described later The high-resolution image ID and high-resolution image data of the high-resolution image including the field in the elapsed days registered by the later-described high-resolution image data selection unit, which is described later. A high-resolution image selection flag indicating whether or not the high-resolution image registered by the selection unit is used in a high-resolution image data analysis unit described later.

  When the low resolution image including the field in the elapsed days cannot be acquired, the vegetation index average value of the record is not registered. Further, when a high-resolution image including the field in the elapsed days cannot be acquired, the high-resolution image ID of the record is not registered. In that case, the high-resolution image selection flag is not registered.

<Estimated growth degree table 11-7>
As shown in FIG. 8, the estimated growth degree table 11-7 is a table in which information on the estimated growth degree of the harvested portion is registered.

For example, the following data is stored.
Cropping ID, estimated growing degree of the planting estimated by the growing degree estimating unit described later.

<Pixel analysis result table 11-8>
The pixel-by-pixel analysis result table 11-8 illustrated in FIG. 9 is a table in which information on the vegetation index by pixel is registered for each cropping, for each pixel, and for each growth stage for high-resolution image data analysis.

For example, the following data is stored.
A cropping ID, a pixel ID for uniquely identifying a pixel in a high-resolution image of the field, and a vegetation index for each pixel analyzed by the high-resolution image data analysis unit for each growth stage to be analyzed for high-resolution image data. When there are a plurality of growth stages subject to high-resolution image data analysis, columns corresponding to the number of vegetation indexes are prepared.

<Estimated normal stock ratio table 11-9>
As shown in FIG. 10, the estimated normal stock rate table 11-9 is a table in which information on the estimated value of the normal stock rate in the field is registered for each crop.

For example, the following data is stored.
Cropping ID, estimated normal stock rate estimated by the normal stock rate estimation unit.

<Predicted yield table 11-10>
As shown in FIG. 11, the predicted yield table 11-10 is a table in which information on the predicted yield is registered for each crop that is a prediction target.

For example, the following data is stored.
A cropping ID, a field ID, a cropping start date, a planned harvest date that is a date on which the crop is scheduled to be harvested, and a predicted harvest amount predicted by a harvest amount prediction unit described later.

<Achievement yield table 11-11>
As shown in FIG. 12, the actual yield table 11-11 is a table in which information on the actual yield in past cropping is registered.

For example, the following data is stored.
The cropping ID, the field ID, the cropping start date, the harvesting date that is the date on which the crop was harvested, and the actual yield that is the actual value of the crop yield.

Next, the overall processing unit and the processing of each processing unit will be described.
≪Description of processing≫
<Overall processing>
FIG. 13 is a flowchart showing the flow of crop yield prediction processing according to the present embodiment. The description will be made with reference to FIG.

(Step S <b> 1) The prediction target cropping setting unit 21-1 sets the cropping ID of the prediction target cropping operated via the input device 3.

(Step S2) The low resolution image data analysis unit 21-2 averages the vegetation index of the field with the prediction target cropping on the image shooting date of the image data from the low resolution image data registered in the low resolution image information table 11-4. The value is calculated and registered in the vegetation-specific vegetation index table 11-6. The average value is used as an example, and a representative value based on other statistical processing may be used. The same applies to the following processing.

(Step S3) The growth model adaptation unit 21-3 adapts the growth model, which is a curve representing the daily change of the vegetation index, to the time-series data of the vegetation index average value, and the estimated vegetation index average value that is the adapted value Is calculated and registered in the vegetation-specific vegetation index table 11-6.

(Step S4) The growth stage estimation unit 21-4 estimates the growth stage of the target crop for each elapsed day from the start date of cropping based on the change rate and / or the number of days of the adapted growth model, and the vegetation index classified by cropping Register in Table 11-6.

(Step S5) The growth degree estimation unit 21-5 acquires the growth stage as the growth degree target from the growth stage table 11-3, and the estimated vegetation index average value of the prediction target cropping from the vegetation-specific vegetation index table 11-6. And the growth stage, and the estimated growth degree of the target planting is calculated by summing the estimated vegetation index average value of the target planting corresponding to the growth stage subject to the growth degree calculation target, and the estimated growth degree table 11- 7 is registered.

(Step S6) The high-resolution image data selection unit 21-6 extracts a high-resolution image including the farm field with the prediction target crop from the high-resolution image registered in the high-resolution image information table 11-5, and the growth stage table 11 Based on the elapsed days selection formula registered in -3 in advance, the high resolution image data to be analyzed is selected for each prediction target cropping and for each growth stage of the high resolution image data analysis target, and the vegetation index table 11-by cropping Register to 6.

(Step S7) The high resolution image data analysis unit 21-7 extracts the image of the field from the high resolution image data to be analyzed, performs coordinate conversion so that the position of the field matches, For each pixel, a multidimensional vector is calculated by combining the vegetation index for each pixel calculated from the high-resolution image data to be analyzed, and registered in the pixel-by-pixel analysis result table 11-8.

(Step S8) The normal stock rate estimation unit 21-8 extracts the pixels of the normal stock portion by classifying the multi-dimensional vectors by unsupervised learning by clustering, and calculates the field from the ratio of the extracted pixels to all the pixels of the field. The normal stock rate is estimated and registered in the estimated normal stock rate table 11-9.
Here, in addition to the normal stock rate or in place of the normal stock rate, the missing stock rate can also be obtained.

(Step S9) The yield predicting unit 21-9 predicts the yield of the crop to be predicted using a yield prediction model including at least the estimated growth degree and the normal stock rate as explanatory variables, and the predicted yield is calculated in the predicted yield table 11. Register at -10.

(Step S <b> 10) The predicted yield output unit 21-10 acquires the predicted yield registered in the predicted yield table 11-10 and displays it on the output device 5.

Next, a processing example in each processing unit will be described in detail.
<Processing in Low Resolution Image Data Analysis Unit 21-2 (FIG. 14)>
Step S2 includes the following processing in detail.

(Step S2-1) The low-resolution image data analysis unit 21-2 obtains the field ID of the prediction target crop and the crop start date D _START from the cropping information table 11-1 using the cropping ID of the prediction target cropping as a key. .

(Step S2-2) The shape file storing the GIS information of the field is acquired from the field-specific GIS information table 11-2 using the field ID of the crop to be predicted as a key. Here, as described above, the shape file is a data format in which the shape of the field can be expressed by polygons, and the position of the field on the map can be expressed by using longitude and latitude as the coordinate system. . By using the shape file, it is possible to determine whether the image of the field is included in the satellite image, or to extract the image of the field from the satellite image.

(Step S2-3) It is determined whether or not the low resolution image data analysis processing has been completed for all the low resolution images registered in the low resolution image information table 11-4. If not completed, the process proceeds to step S2-4. If completed, the processing of the low resolution image data analysis unit 21-2 is terminated.

(Step S2-4) The low resolution image ID of the next low resolution image data registered in the low resolution image information table 11-4 is acquired.

(Step S2-5) The image shooting date D _IMAGE of the image is acquired from the low resolution image information table 11-4 using the low resolution image ID as a key, and D = D _IMAGE -D _START +1 Elapsed days D are calculated.

(Step S2-6) A low resolution image file is acquired from the low resolution image information table 11-4 using the low resolution image ID as a key, and the image of the field is included in the low resolution image file using a shape file. It is determined whether or not.
If the number of days D ≧ 1 from the cropping start date and the image of the field is included in the low-resolution image file (yes), the process proceeds to step S2-7. Otherwise (no), the process returns to step S2-3.

(Step S2-7) The image of the field is extracted from the low-resolution image file using the shape file.

(Step S2-8) A vegetation index is calculated and averaged for each pixel with respect to the image of the field, and a vegetation index average value VI is calculated.

As an example of the vegetation index, NDVI (Normalized Difference Vegetation Index) can be used. When the near-infrared reflectance and visible light red reflectance at the pixel P are ρ _NIR (P) and ρ _RED (P), respectively, NDVI at the pixel P is calculated by the following equation.

Furthermore, when the image of the field is composed of pixels P ₁ ,..., P _N , the vegetation index average value VI is calculated by the following formula.

Other examples of vegetation indices include EVI (Enhanced Vegetation Index) and GRVI (Green-Red Vegetation Index) defined by the following formulas. Here, ρ _GREEN (P) and ρ _BLUE (P) are reflectances of visible light green and visible light blue at the pixel P, respectively, and C ₁ , C ₂ , and L are constants.

  EVI is an index corrected for the effects of the atmosphere and background soil. Compared to NDVI, it is more sensitive to dense vegetation and is less likely to saturate. GRVI is an index calculated from visible light green and visible light red, and is sensitive to the greenness of vegetation.

(Step S2-9) The vegetation index average value VI is registered by adding the cropping ID of the cropping to be predicted and the elapsed days D to the vegetation-specific vegetation index table 11-6.

<Processing in Growth Model Adaptation Unit 21-3 (FIG. 15), (FIG. 20)>
Step S3 includes the following processing in detail.

(Step S3-1) The growth model adaptation unit 21-3 sets a growth model f (X), which is a curve representing daily changes in the vegetation index, via the input device 3. A growth model is used to calculate the estimated average vegetation index across the field on days when low resolution images are not available. The value f (D) in the elapsed days D of the growth model f (X) is set as the estimated vegetation index average value in the entire field in the elapsed days D of the prediction target cropping. As a growth model, for example, a polynomial function such as the following Expression (8) or a Gaussian function of Expression (9) can be used. Here, a, b, and c are parameters of the growth model f (x).

(Step S <b> 3-2) An outlier threshold value θ is set via the input device 3. Here, θ is a constant of 0 or more. An outlier can be defined as a value that deviates greatly from other values in statistics.

(Step S3-3) from planting different vegetation index table 11-6, as a key planting ID of the prediction target crop, age _{D n} and vegetation index average VI _n and the set of time-series data T for the predicted target crop = {(D _n , VI _n )} _{n = 1, 2,.}
Here, N is the number of records of the vegetation index table 11-6 according to cropping including the cropping ID of the crop to be predicted and the record in which the vegetation index average value is registered.

(Step S3-4) The parameters of the growth model f (X) are learned so that the sum of squares of errors between the growth model f (X) and the time series data T is minimized.

(Step S3-5) It is determined whether or not 1 ≦ n ≦ N that satisfies f (D _n ) −VI _n ≧ θ exists. When it exists (yes), it progresses to step S3-6. If it does not exist (no), the process proceeds to step S3-7.

(Step S3-6) The point (D _n , VI _n ) is excluded from the time series data T, 1 is subtracted from N, and the process returns to step S3-4.

Generally observed influence of atmospheric clouds in satellite images, soil moisture, due to the influence of the incident and reflection angle of sunlight is smaller than vegetation index average value to be originally observed as vegetation index average value VI _n May be. If the parameters of the growth model f (X) are learned using such VI _n , the estimated vegetation index average value may be underestimated. For example, as shown in the upper diagram of FIG. 20, if there is an outlier, the estimated vegetation index average value is underestimated. Therefore, as shown in the lower diagram of FIG. 20, when outliers are excluded, the values are close to the vegetation index that should be observed.

Therefore, when the vegetation index average value VI _n is lower than the estimated vegetation index average value f (D _n ) by the threshold θ or more, the point (D _n , VI _n ) is excluded from the learning data as an outlier, and the parameter is learned again. To do. As a result, the estimated vegetation index average value based on the growth model f (X) becomes closer to the vegetation index average value that should not be affected by clouds and should be observed.

(Step S3-7) From the crop information table 11-1, the crop start date D _START and the crop end scheduled date D _{END of the} crop to be predicted are acquired using the crop ID of the crop to be predicted as a key. Also, the elapsed days D are initialized with 1.

(Step S3-8) The value f (D) in the elapsed days D of the growth model f (X) is calculated as the estimated vegetation index average value in the elapsed days D of the planting to be predicted, and in the vegetation index table 11-6 by cropping, The value of the estimated vegetation index average value f (D) is registered by adding the cropping ID of the crop to be predicted and the number of elapsed days D.

(Step S3-9) 1 is added to D.

(Step S3-10) It is determined whether or not D> D _END -D _START +1. If that is right (yes), the process of the growth model adaptation part 21-3 will be complete | finished. Otherwise (no), the process returns to step S3-8.

<Processing in Growth Stage Estimator 21-4, FIG. 21>
As shown in step S4, in the growth stage estimation unit 21-4, from the change rate and / or the number of days of the growth model adapted by the growth model adaptation unit 21-3, as shown in FIG. For each day, the growth stages such as “just after planting”, “vegetation increase”, “vegetation stop”, “vegetation decrease”, etc., representing the stage of planting and growth are estimated. For the division of the growth stage, for example, the following criteria can be used.

The day when the value of the differential df (X) / dX representing the rate of change of the growth model f (X) changes from positive to negative and the value of f (X) changes from increasing to decreasing (hereinafter referred to as “day of maximum growth model”) D _MAX ) can be determined. D _MAX is the number of days counted from the start of cropping (the origin (horizontal axis) in FIG. 21). Hereinafter, the planting start date is the first day, the growth model maximum day is the D _MAX day, and the planting end date is the D _END day. As an example of the growth stage division, a method defined as follows can be considered.

As shown in FIG. 21, the stages 1) to 4) are set as follows.
1) “Immediately after planting” is the first day to the tenth day,
2) “Increase in vegetation” is from day 11 to day (D _MAX- 31),
3) “Vegetation stop” means (D _MAX- 30) day to (D _MAX- 1) day,
4) “Decrease in vegetation” is from D _MAX day to D _END day.

  In the growth stage table 11-3, it is desirable to designate the growth stage at the time when the harvested portion grows as the growth stage that is the target of the growth degree aggregation. Details are described in the explanation section of <Growth estimation unit> below.

  In addition, the growth stage that is the target of high resolution image data analysis in the growth stage table 11-3 needs to have at least one shooting date of the high resolution image in the growth stage. Furthermore, when the influence of weather such as clouds is taken into consideration, it is desirable that the same growth stage includes a plurality of shooting days of high-resolution images. Therefore, a sufficient period of at least about 10 days may be provided according to the imaging period of the satellite.

<Processing in the growth degree estimation unit 21-5, FIGS. 16 and 22>
Step S5 includes the following processing in detail.

  The growth degree estimation unit 21-5 estimates the degree of growth of the harvested portion at the time of harvesting in the prediction target crop for use in yield prediction.

  As a preliminary preparation, the growth stage table 11-3 registers the growth stage at the time when the harvested portion grows as the growth stage subject to growth degree counting. The vegetation of the above-ground part at this time is used for photosynthetic activity for growing the harvested part. Therefore, by accumulating the average value of the vegetation index at this time, the degree of growth of the harvested portion at the time of harvest can be estimated. Therefore, this total value is used as the estimated growth degree of the target crop.

In step S5-1, the growth degree estimation unit 21-5 extracts a record in which the growth degree aggregation target column is “Yes” from the growth stage table 11-3, so that the growth degree accumulation target growth stage is obtained. S ₁ ,..., S _N are acquired. Here, N is the number of extracted records. As an example, as shown in FIG. 22, the growth stage from “vegetation stop” where the vegetation portion on the ground reaches almost the maximum and the growth stops, and the photosynthesis activity is most active, to “vegetation reduction” including the harvest time. Is the target for the growth degree.

In step S5-2, the cropping information table 11-1, as a key planting ID of the prediction target crop, to obtain the planting start date _{D START} and planting scheduled end date _{D END} prediction target crop. Also, the elapsed days D are initialized with 1.

  In step S5-3, the estimated growth degree V is initialized to zero.

  In step S5-4, the estimated vegetation index average value f (D) in the lapsed days D of the prediction target crop and the growth from the vegetation index table 11-6 classified by cropping using the cropping ID of the prediction target crop and the elapsed days D as keys. Stage S (D) is acquired.

か否か、すなわち生育ステージＳ（Ｄ）が生育度集計対象の生育ステージＳ_１，…，Ｓ_Ｎのいずれかであるか否かを判定する。そうであれば（ｙｅｓ）、ステップＳ５−６に進む。そうでなければ（ｎｏ）、ステップＳ５−７に進む。 In step S5-5,

Whether, i.e. growth stage S ₁ of the growth stage S (D) the growth of aggregate _object, ... determines whether either S _N. If so (yes), go to Step S5-6. Otherwise (no), the process proceeds to step S5-7.

  In step S5-6, f (D) is added to the estimated growth degree V.

  In step S5-7, 1 is added to D.

In step S5-8, it is determined whether or not D> D _END -D _START +1 (reach the end date of counting). If so (yes), go to Step S5-9. Otherwise (no), the process returns to step S5-4.

  In step S5-9, the estimated growth degree V is registered by adding the cropping ID of the predicted cropping to the estimated growth degree table 11-7.

<Processing in High Resolution Image Data Selection Unit 21-6, FIGS. 17A, 17B, 23, and 24>
Step S6 includes the following processing in detail.

In step S6-1, the high-resolution image data selection unit 21-6, the cropping information table 11-1, as a key planting ID of the prediction target crop, to obtain the field ID and planting start date _{D START} prediction target crop .

  In step S6-2, the shape file storing the GIS information of the field is acquired from the field-specific GIS information table 11-2 using the field ID of the crop to be predicted as a key. The shape file is as described above.

  In step S6-3, it is determined whether the high resolution image data selection processing has been completed for all the high resolution images registered in the high resolution image information table 11-5. If not completed (no), the process proceeds to step S6-4. If completed (yes), the process proceeds to step S6-8.

  In step S6-4, the high resolution image ID of the next high resolution image data registered in the high resolution image information table 11-5 is acquired.

In step S6-5, using this high and low resolution image ID as a key, the image shooting date D _IMAGE of the image is acquired from the high resolution image information table 11-5, and the date of cropping starts from D = D _IMAGE −D _START +1. The elapsed days D are calculated.

  In step S6-6, a high resolution image file is acquired from the high resolution image information table 11-5 using the high resolution image ID acquired in step S6-4 as a key, and the shape file recorded in step S6-2 is used. Then, it is determined whether or not the field image is included in the high-resolution image file. If D ≧ 1 and the image of the field is included in the high-resolution image file (yes), the process proceeds to step S6-7. Otherwise (no), the process returns to step S6-3.

  In step S6-7, the cropping ID of the crop to be predicted and the elapsed days D are added to the vegetation index table 11-6 by cropping, and the high resolution image ID acquired in step S6-4 is registered. Thereafter, the process returns to step S6-3.

  In step S6-8, it is determined whether or not the following processing has been completed for all the growth stages registered in the growth stage table 11-3. If not completed (no), the process proceeds to step S6-9. If completed (yes), the process of the high resolution image data selection unit 21-6 is terminated.

  In step S6-9, the next growth stage S is acquired from the growth stage table 11-3.

  In step S6-10, if the high resolution image data analysis target column of the record is “yes”, it is determined that the growth stage S is the high resolution image data analysis target, and the process proceeds to step S6-11. If “no”, the process returns to step S6-8.

  In step S6-11, an elapsed day selection formula Q is acquired from the growth stage table 11-3 using the growth stage S as a key.

According to the example of the elapsed days selection formula shown in FIG. 4, as shown in FIGS. 23 and 24, the high resolution image data selection processing is as follows.
a) “Immediately after planting” is selected as one having a high number of elapsed days, and thus becomes a high-resolution image shooting date 1) (S ₁ ).
b) “Increase in vegetation” is selected as the one closest to the elapsed number of days (dotted line in FIG. 24) in the growth stage, and thus becomes the high-resolution image capturing date 4) (S ₂ ).
c) “Vegetation stop” is selected as the one with the latest number of elapsed days, and thus becomes the high-resolution image shooting date 9) (S ₃ ).
Details of the elapsed days selection formula will be described in the <Growth Stage Table> column.

In step S6-12, a record is extracted from the vegetation index table 11-6 according to the cropping using the cropping ID of the crop to be predicted and the growth stage S as keys, and the first elapsed day number D _START (S ), Last elapsed days D _END (S), and elapsed days {D _n (S)} _{n = 1,.} Here, N is the number of records in which the high-resolution image ID is registered among the extracted records. Although not shown in the figure, when N = 0, the following processing is omitted, and the process returns to step S6-8.

In step S6-13, the subscript n of the elapsed days is calculated by n = Q (D _START (S), D _END (S), {D _n (S)} _{n = 1,..., N} ).

In step S6-14, the cropping ID and the elapsed days D _n (S) of the cropping to be predicted are added to the vegetation-specific vegetation index table 11-6, and “Yes” is registered in the column of the high resolution image selection flag. Here, n is a subscript n of the elapsed days calculated in step S6-13.

<Processing in High Resolution Image Data Analysis Unit 21-7, FIG. 18, FIG. 25, FIG. 26>
Step S7 includes the following processing in detail.

  In step S7-1, the high-resolution image data analysis unit 21-7 acquires the field ID of the prediction target cropping from the cropping information table 11-1 using the cropping ID of the prediction target cropping as a key.

  In step S7-2, the shape file storing the GIS information of the field is acquired from the field-specific GIS information table 11-2 using the field ID of the crop to be predicted as a key. The shape file is as described above.

  In step S7-3, it is determined whether or not the following processing has been completed for all the growth stages registered in the growth stage table 11-3. If not completed (no), the process proceeds to step S7-4. If it has been completed (yes), the processing of the high resolution image data analysis unit 21-7 is terminated.

  In step S7-4, the next growth stage S is acquired from the growth stage table 11-3.

  In step S7-5, a record is acquired from the growth stage table 11-3 using the growth stage S as a key. If the high resolution image data analysis target column of the record is “yes”, it is determined that the growth stage S is the high resolution image data analysis target, and the process proceeds to step S7-6. If "no", the process returns to step S7-3.

  In step S7-6, a record is extracted from the vegetation index table 11-6 according to cropping using the cropping ID of the crop to be predicted and the growth stage S as a key, and a record in which the column of the high resolution image selection flag is “Yes”. Extract the high-resolution image ID of the record. Note that, if the record exists, it is uniquely determined by the processing of the high resolution image data selection unit 21-6. If it does not exist, it is assumed that the yield of the crop to be predicted cannot be predicted, and all subsequent processing is interrupted.

  In step S7-7, an image of the field is extracted from the high-resolution image data, and coordinate conversion is executed so that the position and orientation of the field match. FIG. 25 is a diagram showing an outline of the coordinate conversion process. The position and orientation of the field in the high-resolution image data vary depending on the position and angle at the time of capturing the satellite. Therefore, the extracted image of the field is coordinate-transformed by translation, rotation, and enlargement / reduction so that the position and orientation of the field match, and, for example, the highest resolution image 1 and the high resolution image 2 Make the north and the west have the same coordinates. If it is necessary to recalculate pixel values as a result of coordinate transformation, re-sampling is performed by a method such as nearest neighbor.

  Next, as shown in step S7-8, a vegetation index VI (S, P) for each pixel in the pixel P is calculated for each pixel P of the image of the field after the coordinate conversion processing in step S7-7 (step S7-8). (See FIG. 26). Then, the cropping ID of the crop to be predicted and the ID of the pixel P are added, and VI (S, P) is registered in the column of the vegetation index by pixel of the growth stage S in the analysis result table 11-8 by pixel.

  Here, as the vegetation index for each pixel, an index different from the vegetation index used in the low-resolution image data analysis unit 21-2 may be used. In particular, GRVI defined by the formula (7) is an index that easily reacts to the greenness of the leaves, and is suitable for identifying a lacking part. In general, in many cases, visible satellite images can obtain pan-sharpened images with higher resolution than near infrared rays. Since GRVI is an index calculated only from the reflectance of visible light, there is an advantage that analysis can be performed with higher resolution than NDVI using near infrared rays.

<Processing in Normal Stock Rate Estimator 21-8, FIG. 19, FIG. 27A, FIG. 27B>
Step S8 includes the following processing in detail.

In step S8-1, the normal stock rate estimation unit 21-8 extracts a record in which the column of the high-resolution image data analysis target is “Yes” from the growth stage table 11-3, thereby obtaining a high-resolution image. The growth stages S ₁ ,..., S _N to be analyzed are acquired. Here, N is the number of extracted records, and is the number of growth stages to be analyzed for high-resolution image data.

In step S8-2, a record is extracted from the pixel-by-pixel analysis result table 11-8 using the cropping ID of the prediction target cropping as a key, and the pixel {P _m registered in the extracted record from the pixel ID column information } Obtain _{m = 1,.} Here, M is the number of extracted records, and is the number of pixels of the entire image of the field with the prediction target crop extracted from the high resolution image.

  In step S8-3, a record is extracted from the pixel-by-pixel analysis result table 11-8 using the cropping ID of the prediction target cropping as a key, and the N dimension defined by the following formula from the information of the column of each vegetation index by pixel A set X of pixel-specific vegetation index vectors is obtained.

In step S8-4, the vector set X of the N-dimensional vegetation index by pixel is divided into K clusters C ₁ ,..., C _k by a clustering method. Here, K is the number of cluster divisions, and is registered in advance as an integer value of at least 2 or more.

As the clustering method, a fuzzy clustering method such as a Fuzzy c-means method may be used. The output in the case of the Fuzzy c-means method is the center point c _k (k = 1,..., K) of each cluster and the degree of membership w _k (x) of each vector x ∈ X to each cluster C _k . . Here, the degree of membership w _k (x) is a real number from 0 to 1.

As shown in FIG. 27A, in step S8-5, for each cluster C _k (k = 1,..., K), the average of the vectors belonging to cluster C _k is calculated, and the center point c _{k of the} cluster is calculated. calculate. When the number of elements of the cluster C _k is M _k , the cluster center point c _k is calculated by the following equation.

  Note that the Fuzzy c-means method does not need to be calculated because it has already been obtained as an output.

In step S8-6, the center point _c 1 of the cluster, ..., from the conditions for _{c k,} identifies clusters _{C k} of the normal strain pixels.

As an example, a method of identifying a cluster C _k of pixels of normal strains can be used by selecting _ck having the largest vegetation index value in the growth stage of “vegetation stop”.

  In the normal strain, the vegetation index increases until the growth stage of “vegetation stop”, whereas in the lacking part, the vegetation index does not increase from the middle of the growth stage of “vegetation increase”. In addition, the vegetation index of the soil portion where the strain has not been planted continues from the growth stage “immediately after planting”.

From this, it is considered that _ck having the largest value of the vegetation index in the growth stage of “vegetation stop” is the center point of the cluster of normal strain pixels. If the growth stage of “vegetation stop” is _SN , and c _j = (c _1j ,..., C _Nj ) (j = 1,..., K), then the cluster specified when using the above method The subscript k can be calculated by the following formula.

  As described above, in the present embodiment, as shown in FIG. 27A, it is possible to specify a cluster into which pixels of a normal stock portion are classified according to a predetermined condition. Therefore, unsupervised learning is used. However, since normal strains can be accurately estimated, there is an advantage that it is not necessary to prepare teacher data in order to learn a classification model.

  FIG. 27B is a diagram illustrating an example of a case where a missing part portion and a normal strain pixel cannot be identified only by information on a single growth stage.

There is a method for specifying a cluster C _k of normal strain pixels by selecting c _k that satisfies the following condition with respect to the threshold value η set via the input device. Here, η is a positive constant, and S _L (1 ≦ L <N) is a growth stage of “vegetation increase”.

The following (A) or (B) holds for all j = 1,…, K:
(A) c _Nk − η ≧ c _Nj
(B) | c _Nk − c _Nj | ≦ η and c _Lk ≧ c _Lj
This example is suitable for cases where the missing plant part is covered with the leaves of surrounding strains at the time of `` vegetation stop '' and there is no difference in the vegetation index between the missing plant pixel and the normal plant pixel. is there. In such a case, in the previous example, only the vegetation index at the growth stage of “vegetation stop” is used, so that an incorrect cluster may be selected. Therefore, if there is no difference of more than η in the vegetation index value in the growth stage of “vegetation stop”, compare with the value of the vegetation index in the growth stage of “vegetation increase” and select the _ck that maximizes the value. To do.

  As in the example shown in FIG. 27B, by considering the vegetation index in a plurality of growth stages simultaneously using a multidimensional vector, pixels of a missing strain and a normal strain that cannot be identified only by information of a single growth stage, It becomes possible to identify properly.

  In the present embodiment, one normal stock pixel cluster is specified, but two or more clusters may be specified as normal stock pixel clusters.

In step S8-7, it counts the number of elements _{M k} of the specified cluster _{C k,} the estimated normal strain ratio R in the prediction target planting is calculated by R = _{M k} / _M.

In the Fuzzy c-means method, M _k is the sum of the degree of attribution of each vector with respect to the cluster C _k and is calculated by the following equation.

In step S8-8, the estimated normal stock rate R is registered by adding the cropping ID of the predicted cropping to the estimated normal stock rate table 11-9.
It is a normal stock rate estimation device that performs this processing.

<Yield prediction unit 21-9: S9>
In the yield prediction unit 21-9, the estimated growth degree estimated by the growth degree estimation unit 21-5 and the normal stock rate estimated by the normal stock rate estimation unit 21-8 are used as a yield prediction model including explanatory variables. Predict the crop yield. As an example, the yield Y can be predicted with the following yield prediction model.

  The parameters a, b, c, and d of the yield prediction model can be estimated based on the actual yield of cropping that has been carried out in the past. When information on cropping that has been carried out in the past is registered in the cropping information table 11-1, a yield prediction value according to equation (16) is calculated for cropping that has been carried out in the past in the same manner as described above. Can do. By learning the parameters so that the sum of squares of the error between the predicted yield value calculated in this way and the actual yield value registered in the actual yield table 11-11 is minimized, the yield prediction model is obtained. Parameters can be estimated.

  Here, a, b, c, d are parameters of the yield prediction model, N is the number of planted plants for the crop to be predicted obtained from the cropping information table 11-1, and R is a normal strain estimated by the normal stock rate estimation unit Rate, V is the estimated growth degree estimated by the growth degree estimation unit, T is the total value of the average temperature at the nearest observation station obtained from the weather information acquisition means, and S is the weather information acquisition means It is the total value at the growth stage that is the target of the growth rate of the amount of solar radiation obtained from the nearest observation station.

  The future average temperature and the amount of solar radiation can be used by calculating predicted values from long-term forecasts such as one-month forecasts and three-month forecasts obtained by the weather information obtaining means 11 and past weather observation results. it can. An example of the calculation method of the predicted value is shown below.

If the long-term forecast is a forecast of “low (small)”, for example, the observation values for the past 30 years are arranged in ascending order, and the first to tenth average values are taken in order from the smallest.
If the forecast is “normal”, the 11th to 20th average values are taken in ascending order.
If the forecast is “higher than normal (more)”, the 21-30th average value is taken in ascending order.
The obtained predicted yield value Y is stored in the predicted yield table 11-10.

<Predicted yield output unit 21-10: S10>
The predicted yield output unit 21-10 can output the predicted yield to the output device 5 based on the predicted yield table 11-10.

  The actual yield table 11-11 shown in FIG. 12 is a table in which information on the actual yield in past cropping is registered.

  For example, the cropping ID, the field ID, the cropping start date, the harvesting date, which is the date when the crop was harvested, and the actual yield, which is the actual value of the crop, are compared with the predicted yield in FIG. And so on.

  As described above, according to the embodiment of the present invention, since there is a yield predicting unit that predicts the yield using the normal stock rate estimated by the normal stock rate estimating unit as an explanatory variable, Therefore, it is possible to make a prediction in consideration of a decrease in the yield due to the above, and to predict the yield with high accuracy.

  Further, since the normal stock rate estimation unit can estimate the normal stock rate using unsupervised learning by clustering, it is not necessary to prepare teacher data to learn the classification model. Therefore, there is an advantage that it can be carried out without conducting prior field surveys and satellite image analysis.

  The above processing and control can be performed by software processing using a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), or FPGA (Field Programmable Gate) hardware.

  In the above-described embodiment, the configuration and the like illustrated in the accompanying drawings are not limited to these, and can be changed as appropriate within the scope of the effects of the present invention. In addition, various modifications can be made without departing from the scope of the object of the present invention.

  Each component of the present invention can be arbitrarily selected, and an invention having a selected configuration is also included in the present invention.

  For example, the present invention can also be used in a normal strain estimation technique that is a basis for yield prediction.

  In addition, a program for realizing the functions described in the present embodiment is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed to execute processing of each unit. May be performed. The “computer system” here includes an OS and hardware such as peripheral devices.

  Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.

  The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory in a computer system serving as a server or a client in that case, and a program that holds a program for a certain period of time are included. The program may be a program for realizing a part of the above-described functions, or may be a program that can realize the above-described functions in combination with a program already recorded in a computer system. At least a part of the functions may be realized by hardware such as an integrated circuit.

  The present invention can be used as a crop yield prediction apparatus for agricultural products.

A ... Yield prediction device, 1 ... Database group (auxiliary storage device), C ... Various interface units, 3 ... Input device, 5 ... Output device, 7 ... Central processing unit (CPU) (control unit), 11 ... Weather information Acquisition means, 21 ... main memory, 21-1 ... prediction target cropping setting unit, 21-2 ... low resolution image data analysis unit, 21-3 ... growth model adaptation unit, 21-4 ... growth stage estimation unit, 21- 5 ... Growth degree estimation unit, 21-6 ... High resolution image data selection unit, 21-7 ... High resolution image data analysis unit, 21-8 ... Normal stock rate estimation unit, 21-9 ... Yield prediction unit, 21- 10: Predicted yield output unit.

Claims (7)

A normal stock rate estimation device for estimating the normal stock rate of agricultural products,
A crop ID that uniquely identifies the crop cropping, a field ID that uniquely identifies the crop field, a crop start date that is the date that the crop was started, and the crop will be harvested and cropping will be terminated. The planting information in which the planned planting end date, which is the date, is registered,
Field-specific GIS information in which information representing the position and shape of the field is registered for each field,
Growth stage information in which a growth stage indicating the cropping and growing stage of the crop and information indicating whether it is a growth stage of image data analysis target are registered,
An image ID for uniquely identifying an image of the first resolution, an image shooting date that is a date when the image of the first resolution is shot, and an image data body of the first resolution are registered. Image information of one resolution,
An image ID for uniquely identifying an image of the second resolution, an image shooting date that is a date when the image of the second resolution is shot, and an image data main body of the second resolution are registered. Two-resolution image information, and a storage device for storing the image information,
A prediction target cropping setting unit for setting a crop target ID of the prediction target from the input device;
Based on the set crop ID, the first resolution image data acquired with reference to the crop information and the first resolution image information is used to predict the crop field to be predicted on the image capturing date. An image data analysis unit for calculating a vegetation index representative value;
With reference to the growth stage information, a growth model adapting unit that calculates an estimated vegetation index representative value by adapting the growth model of the crop to the time series data of the vegetation index representative value;
A growth stage estimation unit that estimates a growth stage for each elapsed day from the start date of cropping, from at least one of the change rate and the number of days of the growth model of the adapted crop,
For each growth stage of the image data analysis target, an image data selection unit for the second resolution for selecting the image data of the second resolution to be analyzed, and a pixel of the farm field with the prediction target in the image of the second resolution For each, the image data analysis unit for the second resolution to calculate a multidimensional vector by combining the vegetation index by pixel calculated from the image data of the second resolution to be the analysis target,
The multi-dimensional vector is classified by clustering based only on the vector value, and one or more clusters in which the pixels of the normal stock portion are classified are identified by a predetermined condition from the obtained clusters, and the identification for all the pixels in the field is performed. A normal stock rate estimation apparatus comprising: a normal stock rate estimation unit that estimates a normal stock rate in a field from a ratio of pixels belonging to a cluster obtained.   The normal stock rate estimation apparatus according to claim 1, wherein the first resolution is lower than the second resolution. The normal stock rate estimator is
The normal stock rate estimation apparatus according to claim 1 or 2, wherein a cluster into which pixels of a normal stock portion are classified is specified from a condition with respect to a center point of the cluster.   The normal stock rate estimation apparatus according to any one of claims 1 to 3, wherein the vegetation index representative value and the vegetation index by pixel are each NDVI, EVI, or GRVI. A yield predicting device for predicting a yield of an agricultural crop, wherein the normal stock rate estimating device according to any one of claims 1 to 4,
A yield prediction apparatus comprising: a yield prediction model that predicts a yield of a crop to be predicted with a yield prediction model including at least the normal stock ratio as an explanatory variable. In the growth stage information, information indicating whether or not the growth stage is a growth stage subject to growth degree counting is registered,
further,
A growth degree estimation unit that calculates the estimated growth degree of the predicted target crop by summing the estimated vegetation index representative values of the predicted target crop in the growth stage of the growth degree target,
The said harvest amount prediction part predicts the harvest amount of cropping to be predicted based on at least the estimated growth degree and the yield prediction model including the normal stock rate as explanatory variables. Yield prediction device. A normal stock rate estimation method for estimating the normal stock rate of agricultural products,
A crop ID that uniquely identifies the crop cropping, a field ID that uniquely identifies the crop field, a crop start date that is the date that the crop was started, and the crop will be harvested and cropping will be terminated. The planting information in which the planned planting end date, which is the date, is registered,
Field-specific GIS information in which information representing the position and shape of the field is registered for each field,
Growth stage information in which a growth stage indicating the cropping and growing stage of the crop and information indicating whether it is a growth stage of image data analysis target are registered,
An image ID for uniquely identifying an image of the first resolution, an image shooting date that is a date when the image of the first resolution is shot, and an image data body of the first resolution are registered. Image information of one resolution,
An image ID for uniquely identifying an image of the second resolution, an image shooting date that is a date when the image of the second resolution is shot, and an image data main body of the second resolution are registered. Referring to each information of the storage device storing the image information of the second resolution,
A prediction target cropping setting step for setting the cropping ID of the prediction target from the input device;
Based on the set crop ID, the first resolution image data acquired with reference to the crop information and the first resolution image information is used to predict the crop field to be predicted on the image capturing date. An image data analysis step for calculating a vegetation index representative value;
With reference to the growth stage information, a growth model fitting step of calculating an estimated vegetation index representative value by fitting a growth model of the crop to the time series data of the vegetation index representative value;
A growth stage estimation step for estimating a growth stage for each elapsed day from the start date of cropping, from at least one of the change rate and the number of days of the growth model of the adapted crop,
An image data selection step for selecting the second resolution image data to be analyzed for each growth stage of the image data analysis target, and a pixel of the farm field with the prediction target in the second resolution image Image data analysis step for the second resolution, which combines a vegetation index for each pixel calculated from the image data of the second resolution to be analyzed, and calculates a multidimensional vector,
The multi-dimensional vector is classified by clustering based only on the vector value, and one or more clusters in which the pixels of the normal stock portion are classified are identified by a predetermined condition from the obtained clusters, and the identification for all the pixels in the field is performed. A normal stock rate estimation step for estimating the normal stock rate in the field from the proportion of pixels belonging to the selected cluster,
A normal stock rate estimation method comprising:

Images