WO2023074466A1

WO2023074466A1 - Machine learning device, estimation device, and program

Info

Publication number: WO2023074466A1
Application number: PCT/JP2022/038750
Authority: WO
Inventors: 克彦近藤; 昭彦松村
Original assignee: Ｔｅｓｎｏｌｏｇｙ株式会社
Priority date: 2021-10-29
Filing date: 2022-10-18
Publication date: 2023-05-04
Also published as: JP2023066966A

Abstract

A machine learning device according to the present invention comprises: a storage unit that stores soil data of cultivated land where plants of a prescribed kind are cultivated, image features extracted from captured images of the cultivated land, and growth data of the roots of the plants of the prescribed kind and/or the amounts of carbon dioxide absorption of the roots of the plants of the prescribed kind at the times when the images were captured; and a model generation unit that generates, through machine learning, an estimation model for estimating, from soil data and image features extracted from a captured image, growth data of the roots of a plant of the prescribed kind and/or the amount of carbon dioxide absorption of the roots of the plant of the prescribed kind at the time when the image was captured.

Description

Machine learning device, estimation device and program

The present invention relates to technology for estimating plant growth.

With the recent global warming, tree planting activities are widely carried out to reduce the amount of carbon dioxide in the atmosphere through plant photosynthesis. In addition, it is known that the roots of plants have the ability to fix carbon, and we believe that accurately determining the length of the roots of plants planted in the field will be important in estimating the amount of carbon absorbed. It is

In Patent Document 1, effective and objective rhizosphere elements in soil for quantitatively grasping the distribution amount and dynamics of fine roots such as trees are automatically and non-destructively measured without manual work such as visual inspection. is disclosed.

Japanese Patent No. 5078508 WO2019/097892

However, the technology disclosed in Patent Document 1 is a technology for predicting the growth situation in a specific field, and when targeting a wide range of fields, there is a problem that the introduction of the mechanism requires a great deal of cost.

Further, in Patent Document 2, using an aerial image or the like, the photographed images of a field are classified based on the photographing altitude, and from the classified image group, the field, the division of the field, and the plants in the field, Techniques for reconstructing 3D image data and the like and converting them into visualization data based on the field conditions or the imaging conditions have been disclosed.

However, in the technique disclosed in Patent Document 2, it is necessary to prepare images taken at different altitudes for each section, and a UAV (Unmanned Aerial Vehicle) such as a drone is flown multiple times to one field. There is a need.

In order to solve this problem, there is a need for a technology to grasp the amount of carbon dioxide absorbed by plants cultivated in a given area, and by extension, the amount of carbon dioxide absorbed by the roots, based on images taken by satellites, without using a large-scale system. It is

The present invention is an invention completed based on the recognition of the above problems, and its main purpose is to provide a technique that can easily estimate the state of the roots of plants.

A machine learning device in one aspect of the present invention comprises:
At least one of soil data of a field in which a given plant is cultivated, an image feature amount extracted from a photographed image of the field, root growth data of the given plant at the time of photographing the photographed image, and carbon dioxide absorption amount of the root of the given plant. a storage unit that stores and
Using machine learning to generate an estimation model for estimating at least one of the root growth data of a given plant and the amount of carbon dioxide absorbed by the roots of a given plant at the time of photographing from the soil data and the image feature quantity extracted from the photographed image. and a model generator.

An estimating device in one aspect of the present invention comprises:
a storage unit that stores soil data of a field in which a predetermined plant is cultivated;
an image acquisition unit that acquires a photographed image of a field;
an extraction unit that extracts an image feature amount from the captured image;
Soil data of a field in which a predetermined plant was cultivated in the past and an image feature amount extracted from a photographed image of the field in the past cultivation are used as input data, and the root of the predetermined plant at the time of photographing the photographed image in the past cultivation. Image feature values extracted from stored soil data and captured images using an estimation model machine-learned using at least one of the growth data and the carbon dioxide absorption amount of the roots of a predetermined plant as output data and an estimating unit for estimating at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the acquired captured image.

A machine learning device in another aspect of the present invention comprises:
Soil data of a field for cultivating a predetermined plant that can be cultivated in the second period with the roots remaining after the first harvesting, an image feature amount extracted from the first captured image of the field in the first period, At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the first photographing of one photographed image, and the image feature amount extracted from the second photographed image of the field in the second period; a storage unit that stores at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of the second photographing of the second photographed image;
Machine learning a first estimation model for estimating at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of the first photographing from the soil data and the image feature amount of the first photographed image. Generated by
Machine learning a second estimation model for estimating at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of the second photographing from the soil data and the image feature amount of the second photographed image. and a model generation unit that generates by

An estimating device in another aspect of the present invention comprises:
a storage unit for storing soil data of a field for cultivating a predetermined plant that can be cultivated for a second period with roots remaining after the first harvest;
an image acquisition unit that acquires a first captured image of the field in the first period or a second captured image of the field in the second period;
an extraction unit that extracts an image feature amount from the first captured image or the second captured image;
When the first photographed image is acquired, the soil data of the field where the predetermined plant was cultivated in the past and the image feature amount extracted from the photographed image of the field in the first period of past cultivation are used as input data, Using a first estimation model machine-learned using at least one of the root growth data of a predetermined plant and the amount of carbon dioxide absorbed by the root of a predetermined plant at the time of photographing the photographed image in the first period in the past cultivation as output data, At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the first captured image, based on the stored soil data and the image feature amount extracted from the first captured image. , and
When the second photographed image is acquired, the soil data of the field where the predetermined plant was cultivated in the past and the image feature amount extracted from the photographed image of the field in the second period of past cultivation are used as input data, Using a second estimation model machine-learned using at least one of the root growth data of a predetermined plant and the amount of carbon dioxide absorbed by the roots of a predetermined plant at the time of capturing the captured image in the second period in the past cultivation as output data, At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the second captured image, based on the stored soil data and the image feature amount extracted from the second captured image. an estimating unit for estimating

A machine learning device in yet another aspect of the present invention comprises:
Soil data of a field for cultivating a predetermined plant that can be cultivated in the second period with the roots remaining after the first harvesting, an image feature amount extracted from the first captured image of the field in the first period, At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the first photographing of one photographed image, and the image feature amount extracted from the second photographed image of the field in the second period; At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the second photographing of the second photographed image, and the root growth data of the predetermined plant and the roots of the predetermined plant at the end of the first period a storage unit that stores at least one of the carbon dioxide absorption amount of
Machine learning a first estimation model for estimating at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of the first photographing from the soil data and the image feature amount of the first photographed image. Generated by
Based on the soil data, the image feature amount of the second photographed image, and at least one of the root growth data of the predetermined plant at the end of the first period and the carbon dioxide absorption amount of the roots of the predetermined plant, the predetermined plant at the time of the second photographing. and a model generation unit that generates a second estimation model for estimating at least one of the root growth data of the plant and the carbon dioxide absorption amount of the root of the predetermined plant by machine learning.

An estimating device according to yet another aspect of the present invention comprises:
a storage unit for storing soil data of a field for cultivating a predetermined plant that can be cultivated for a second period with roots remaining after the first harvest;
an image acquisition unit that acquires a first photographed image of the field at the end of the first period and further acquires a second photographed image of the field in the second period;
an extraction unit that extracts an image feature amount from the first captured image and further extracts an image feature amount from the second captured image;
Using as input data soil data of a field in which a predetermined plant was cultivated in the past and an image feature amount extracted from a photographed image of the field in the first period of past cultivation, a photographed image of the first period of past cultivation. Using a first estimation model machine-learned using at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the root of the predetermined plant at the time of photographing as output data, the stored soil data and the first estimating at least one of the growth data of the roots of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the end of the first term from the image feature amount extracted from the photographed image;
Soil data of a field where a predetermined plant was cultivated in the past, image feature values extracted from a photographed image of the field in the second period of the past cultivation, and roots of the predetermined plant at the end of the first period of the past cultivation. and at least one of the carbon dioxide absorption amount of the roots of the predetermined plant is used as input data, and the growth data of the roots of the predetermined plant and the carbon dioxide of the roots of the predetermined plant at the time of capturing the captured image in the second period in the past cultivation Using a second estimation model machine-learned with at least one of the carbon uptake as output data, the stored soil data, the image feature amount extracted from the second captured image, and the estimated end of the first term The root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the second captured image, based on at least one of the root growth data of the predetermined plant at the time and the amount of carbon dioxide absorbed by the roots of the predetermined plant. and an estimating unit that estimates at least one of

According to the present invention, the state of plant roots can be easily estimated.

FIG. 4 is a diagram showing an overview of the test phase; It is a figure which shows the growth period of 3rd period cultivation. It is a figure which shows the outline|summary of an operation stage. It is a figure which shows the phase of information processing. It is a functional block diagram of a growth management apparatus. It is a data structure diagram of a soil data storage unit. 4 is a data structure diagram of feature data; FIG. 4 is a data structure diagram of root data; FIG. Fig. 10 is a flow chart showing the processing steps in the test data collection phase; 1 is a configuration diagram of a neural network in an embodiment; FIG. FIG. 4 is a data structure diagram of teacher data for the first period in the embodiment. FIG. 10 is a data structure diagram of teacher data for the second term in the embodiment. 4 is a flow chart showing a process in an estimation model generation phase; FIG. 10 is a diagram showing a process in an estimation model usage phase; FIG. 10 is a configuration diagram of a neural network in modification 1; FIG. 11 is a data structure diagram of teacher data for the second term in modification 1; FIG. 11 is a data structure diagram of teacher data for the third period in modification 1; It is a block diagram of a growth management system.

[Embodiment]
Biomass, which is carbon neutral, is expected as a measure to reduce carbon dioxide. One example of biomass, sorghum, becomes an ethanol-producing material. Sorghum is a plant of the grass family. Sorghum grown on one hectare is said to absorb 90 tons of carbon dioxide through its leaves and stems, and 60 tons through its roots. In particular, roots play a role in retaining carbon dioxide in the soil.

Here, we provide a mechanism for understanding the amount of carbon dioxide absorbed during the period from sorghum cultivation to harvest. A satellite image of a field where sorghum is cultivated is taken from above, and the captured satellite image is used. In addition, artificial intelligence technology is used to estimate the growth rate of sorghum during cultivation. In particular, measure root growth. In this example, in order to perform supervised learning, it is assumed that experimental cultivation is performed in order to collect data serving as the basis for teacher data.

　Sorghum is cultivated in triplicate. In the second period, the grass of the second period is grown from the roots left after cutting the grass in the first period. Furthermore, in the third period, the grass of the third period is grown from the roots left after cutting the grass in the second period. In other words, the grass is grown three times while the roots are stretched.

FIG. 1 is a diagram showing an overview of the testing stage.
Field 220a, field 220b and field 220c cultivate sorghum in soil with different characteristics. In order to know the degree of growth, the artificial satellite 302 photographs the farm field 220a, the farm field 220b, and the farm field 220c from above. The captured satellite image is transmitted to the image providing server 300 . The growth management device 100 acquires satellite images of the farm field 220 a , the farm field 220 b and the farm field 220 c from the image providing server 300 . The growth management device 100 is a device used for managing the growth of a given plant, for example, in

fields

220a, 220b and 220c.

The soil data of the farm field 220a, the farm field 220b, and the farm field 220c are sent from the farm field terminal 200 to the growth management device 100 before the start of cultivation. Soil data is data representing characteristics of soil, such as acidity, drainage, fertility, and viscosity.

Also, during cultivation, data representing the state of root growth (hereinafter referred to as "root growth data") is sent from the field terminal 200 to the growth management device 100. Root growth data are, for example, root length, root weight, root volume, and the like.

Soil characteristics affect how sorghum grows. In particular, the way the roots spread is easily affected by the soil. In other words, there is a relationship between soil data and root growth data. In this example, it is assumed that test cultivation was performed in 2020 in field 220a, field 220b, and field 220c, and basic data was obtained.

The growth management device 100 performs machine learning using the data obtained in the test stage to generate an estimation model for estimating the root growth state.

FIG. 2 is a diagram showing the growth period of three-season cultivation.
The growth of a given plant (for example, sorghum) that can be cultivated in the second and third periods with the roots left after the first harvest will be described.

In the first period, we will start with seeding and cultivate until the 70th day, that is, the 10th week. Meanwhile, satellite images are acquired every seven days. Root measurement is performed at that timing. After the ninth week, the fruit is formed and the leaves and stems are cut. Some grass is grown as is for 10 week photography and measurements. In Modified Example 1, which will be described later, a satellite image is acquired even when the first harvest is performed, and the roots are measured.

The first harvest marks the beginning of the second term. The second period is also to be cultivated until the 10th week. In addition, satellite images are acquired every seven days, and root measurements are performed. After the ninth week, the fruit is formed and the leaves and stems are cut. Some grass is grown as is for 10 week photography and measurements. In Modified Example 1, which will be described later, a satellite image is acquired and roots are measured even when the plant is pruned for the second time.

The second harvest marks the beginning of the third term. The 3rd term shall also be cultivated until the 10th week. In addition, satellite images are acquired every seven days, and root measurements are performed. After the ninth week, the fruit is formed and the leaves and stems are cut. Some grass is grown as is for 10 week photography and measurements. Roots remain in the ground.

FIG. 3 is a diagram showing an overview of the operation stage.
In this example, it corresponds to the year 2021. Soil data is sent from the farm field terminal 200 to the growth management apparatus 100 also in the operation stage. Also, as described in connection with FIG. 2, satellite images are similarly acquired in each week of each period. However, in the operation stage, it is not necessary to measure the roots. This is because the growth management apparatus 100 can obtain the root growth data and the amount of carbon dioxide absorbed by the root using the estimation model.

The amount of carbon dioxide absorbed by roots (per plant) can be obtained by root weight x predetermined magnification and root volume x predetermined magnification. Since the amount of carbon dioxide uptake by roots is correlated with the length of roots, the amount of carbon dioxide uptake by roots may be obtained from a correspondence table or formula showing the correlation between root length and carbon dioxide uptake by roots. . The amount of carbon dioxide absorbed by roots (per unit area) can be obtained by multiplying the amount of carbon dioxide absorbed by roots (per plant)×grass density (plant/unit area). A unit area is, for example, a hectare.

FIG. 4 is a diagram showing the phases of information processing.
In the test data collection phase (S10), performance data in the test stage (Fig. 1) is collected. In the estimation model generation phase (S12), a learning model is generated using the collected performance data as teacher data. The learning model in this example is called an estimation model. In the estimation model utilization phase (S14), the generated estimation model is used to obtain root growth data and root carbon dioxide absorption in the operation stage (FIG. 3).

Since root growth and carbon dioxide absorption differ depending on the type of plant, the growth characteristics of specific plants are reflected in the estimation model. For sorghum cultivation, an inference model for sorghum is generated. For other plants, an inference model is generated for that plant.

In addition, the grass grows differently in the first, second, and third periods. Therefore, separate estimation models are generated in which each growth mode is reflected. The estimation model for the first period is called a "first estimation model". The estimation model for the second period is called a "second estimation model". The estimation model for the third period is called a "third estimation model". In this example, a neural network is used for machine learning. Machine learning processing using neural networks may be general. Also, means other than neural networks, such as natural language processing represented by morphological analysis, may be used for machine learning.

FIG. 5 is a functional block diagram of the growth management device 100. As shown in FIG.
The growth management apparatus 100 has the function of a machine learning device that performs processing in the test data collection phase (S10) and the estimation model generation phase (S12) and generates an estimation model by machine learning. Furthermore, in the estimation model utilization phase (S14), the growth management device 100 has a function of an estimation device that obtains satellite imagery and obtains root growth data and root carbon dioxide absorption using the estimation model.

Each component of the growth management apparatus 100 includes computing units such as a CPU (Central Processing Unit) and various coprocessors, storage devices such as memory and storage, and hardware including wired or wireless communication lines connecting them. , is stored in a storage device and implemented by software that supplies processing instructions to the computing unit. A computer program may consist of a device driver, an operating system, various application programs located in their higher layers, and a library that provides common functions to these programs. Each illustrated block mainly indicates a functional unit block. Each block may be implemented by causing a computer to execute a program stored in a storage device.

The growth management device 100 includes a user interface processing unit 102, a data processing unit 104, a communication processing unit 108 and a data storage unit 106. The user interface processing unit 102 receives user operations via a mouse or a touch panel, and is in charge of user interface processing such as image display and audio output. A communication processing unit 108 is in charge of communication processing via a network, short-range wireless communication, or the like. A data storage unit 106 stores various data. The data processing unit 104 executes various processes based on data acquired by the user interface processing unit 102 and the communication processing unit 108 and data stored in the data storage unit 106 . The data processing unit 104 also functions as an interface for the user interface processing unit 102 , the communication processing unit 108 and the data storage unit 106 .

The user interface processing unit 102 has an input unit 122 for inputting data by user operation and an output unit 120 for outputting data to be provided to the user.

The communication processing unit 108 includes a transmitting unit 180 that transmits various data and a receiving unit 182 that receives various data.

The data processing unit 104 includes a soil data acquisition unit 140 , an image acquisition unit 142 , a feature amount extraction unit 144 , a root data acquisition unit 146 , a teacher data generation unit 148 , an estimation model generation unit 150 and a root growth estimation unit 154 .
The soil data acquisition unit 140 acquires soil data from the field terminal 200 using the transmission unit 180, for example. The soil data acquisition unit 140 may acquire soil data received by the input unit 122 .

The image acquisition unit 142 acquires a photographed image of the field 220 (a satellite image in this example). In this example, the image acquiring unit 142 transmits a satellite image request specifying the position of the field 220 and the shooting date from the transmitting unit 180 to the image providing server 300, and the receiving unit 182 receives the captured satellite image. receive. Specifically, in the test stage and the operation stage, the satellite that captured the field 220 at the timing of each week of the first term, each week of the second term, and each week of the third term of the field 220a, the field 220b, and the field 220c An image is acquired.

The feature quantity extraction unit 144 extracts the feature quantity from the acquired photographed image of the farm field 220 (satellite image in this example). The growth management apparatus 100 uses feature amounts (hereinafter referred to as “image feature amounts”) extracted from satellite images. The image feature amount is, for example, the average color of the field range, the percentage of green in the field range, the degree of green density in the field range, and the height of the plant. As plants grow, leaves and stems grow larger, so the average color of the field area changes from earthy to green. In addition, as the plants grow, the proportion of green in the field area increases, and the degree of green density also increases. Furthermore, as the plant grows, the height of the plant also increases. In this way, as the leaves and stems grow, the roots grow as well. Therefore, the image feature amount has a correlation with the root growth data. In addition, the image features are also correlated with root carbon dioxide uptake associated with root growth data. Specifically, the feature amount extraction unit 144 extracts the captured image acquired in the first period (hereinafter referred to as “first captured image”), the captured image acquired in the second period (hereinafter referred to as “second captured image”). ”) and the captured image acquired in the third period (hereinafter referred to as “third captured image”). For example, when pixel colors are expressed in RGB format, the R value of the average color of the field range is the average of the R values of the pixels included in the field range, and the G value of the average color of the field range is the average value of the pixels included in the field range. It is the average value of the G values of the included pixels, and the R value of the average color of the field range is the average value of the R values of the pixels included in the field range. Average color is an example of an index that indicates the color characteristics of an image. Instead of the average color, it is also possible to use an index (an index indicating the color feature of an image) that is patterned by a method other than averaging, such as a color tone pattern or contrast.

The root data acquisition unit 146 acquires root growth data from the field terminal 200 . The receiving unit 182 may receive the root growth data through network communication such as the Internet or LAN (Local Area Network), or may receive the root growth data through short-range wireless communication. Alternatively, the root data acquisition unit 146 may acquire root growth data received by the input unit 122 .

The teacher data generation unit 148 generates teacher data used for machine learning based on soil data, image feature values, and root data (at least one of root growth data and root carbon dioxide absorption). The estimation model generator 150 includes a learning engine 152 . The estimation model generation unit 150 uses the learning engine 152 to generate a learning model (estimation model) based on teacher data. When the learning model uses a neural network, it has an input node corresponding to each index of the soil data and the image feature value which are the input variables, an output node corresponding to the index of the root data which is the output variable, and an intermediate node. A neural network is used to generate weight data that indicates the strength of connections between nodes. A configuration example of the neural network will be described later with reference to FIG. In addition, when using a neural network, each index value of the soil data and the image feature value of the input variables prepared as application data is set to the input node corresponding to the index, and output based on the weight data An index value of root data is obtained from the node.

In other words, the estimation model generation unit 150 generates the soil data and the image feature amount (in this example, the average color of the field range) extracted from the photographed image (in this example, the satellite image). machine learning to generate an estimation model for estimating at least one of the root growth data of a given plant (sorghum in this example) and the amount of carbon dioxide absorbed by the given plant's roots. More specifically, for the first period, the estimation model generation unit 150 generates root growth data and A first estimation model for estimating at least one of carbon dioxide uptake by roots of a given plant is generated by machine learning. In addition, regarding the second term, the estimation model generation unit 150 calculates the root growth data of the predetermined plant and the root growth data of the predetermined plant at the time of capturing the second captured image from the soil data and the image feature amount of the second captured image. A second estimation model for estimating at least one of carbon dioxide absorption is generated by machine learning. Furthermore, regarding the third term, the estimation model generation unit 150 calculates the root growth data of the predetermined plant and the root growth data of the predetermined plant at the time of capturing the third captured image based on the soil data and the image feature amount of the third captured image. A second estimation model for estimating at least one of carbon dioxide absorption is generated by machine learning.

The root growth estimation unit 154 estimates root growth data in the estimation model use phase (S14). Root growth estimation unit 154 includes estimation model utilization unit 156 . The estimation model utilization unit 156 utilizes the estimation model for estimating root growth data. The estimation model using unit 156 applies the soil data of the farm field 220 to be estimated and the image feature amount extracted from the acquired photographed image (satellite image in this example) to the estimation model as input variables.

That is, the root growth estimating unit 154 collects the soil data of the field 220 in which the predetermined plant (sorghum in this example) was cultivated in the past cultivation (the test stage in this example), and the photographed image of the field 220 in the past cultivation ( In this example, image feature values extracted from satellite images) are used as input data, and at least one of root growth data of a predetermined plant at the time of photographing images in past cultivation and carbon dioxide absorption amount of the roots of the predetermined plant. is used as the output data and a machine-learned estimation model is used. Then, the root growth estimating unit 154 uses the soil data of the target field stored in the soil data storage unit 160 and the image feature amount extracted from the photographed image acquired by the image acquiring unit 142 to obtain the photographed image. At least one of the growth data of the roots of the predetermined plant and the carbon dioxide absorption amount of the roots of the predetermined plant at the time of photographing is estimated.

More specifically, the root growth estimating unit 154, when the first photographed image is acquired in the first period of the operation stage, determines that the predetermined plant (in this example, Soil data of a field 220 in which sorghum was cultivated and an image feature amount extracted from a photographed image of the field 220 in the first period of past cultivation are used as input data, and a photographed image of the first period of past cultivation is taken. A first estimation model machine-learned using at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the root of the predetermined plant at the time point as output data is used. Then, the root growth estimating unit 154 calculates, from the soil data of the target field stored in the soil data storage unit 160 and the image feature amount extracted from the first captured image, the predetermined plant estimating at least one of root growth data of plants and carbon dioxide uptake of roots of a given plant.

Further, when the root growth estimation unit 154 acquires the second photographed image in the second term of the operation stage, the predetermined plant (sorghum in this example) was cultivated in the past cultivation (test stage in this example). The soil data of the field 220 and the image feature amount extracted from the photographed image of the field in the second period in the past cultivation are used as input data, and the root of the predetermined plant at the time of photographing the photographed image in the second period in the past cultivation. A second estimation model machine-learned using at least one of the growth data and the carbon dioxide absorption amount of the roots of a predetermined plant as output data is used. Then, the root growth estimating unit 154 calculates, from the soil data of the target field stored in the soil data storage unit 160 and the image feature amount extracted from the second captured image, estimating at least one of root growth data of plants and carbon dioxide uptake of roots of a given plant.

In addition, when the root growth estimating unit 154 acquires the third photographed image in the third term of the operation stage, the root growth estimation unit 154 cultivates the predetermined plant (sorghum in this example) in the past cultivation (test stage in this example). The soil data of the field 220 and the image feature amount extracted from the photographed image of the field in the third period in the past cultivation are used as input data, and the image of the predetermined plant at the time of photographing the photographed image in the third period in the past cultivation. A third estimation model machine-learned using at least one of the root growth data and the carbon dioxide absorption amount of the roots of the predetermined plant as output data is used. Then, the root growth estimating unit 154 uses the soil data of the target field stored in the soil data storage unit 160 and the image feature amount extracted from the third photographed image to determine whether the predetermined plant at the time of photographing the third photographed image. estimating at least one of root growth data of plants and carbon dioxide uptake of roots of a given plant.

Data storage unit 106 includes soil data storage unit 160 , image storage unit 162 , feature amount data storage unit 164 , root data storage unit 166 , teacher data storage unit 168 and estimation model storage unit 170 .
The soil data storage unit 160 stores soil data. The soil data storage unit 160 will be described later with reference to FIG. The image storage unit 162 stores a photographed image (a satellite image in this example) in association with a combination of year, field 220, and photographing date and time. It should be noted that the shooting date and time can specify the period and week of each field 220 . The feature amount data storage unit 164 stores feature amount data. The feature amount data will be described later with reference to FIG. The root data storage unit 166 stores root data. The configuration of the root data storage unit 166 will be described later with reference to FIG. The teacher data storage unit 168 stores teacher data. The teacher data will be described later with reference to FIGS. 11 and 12. FIG. The estimated model storage unit 170 stores an estimated model for each period in each field 220 . The data storage unit 106 includes a field data storage unit (not shown) that stores field data such as the geographical position and range of each field 220, and the density (strain/unit area) of a given plant (in this example, sorghum). .

That is, the data storage unit 106 stores the soil data of the field 220 where a predetermined plant (sorghum in this example) is cultivated, the image feature amount extracted from the photographed image of the field 220 (satellite image in this example), and the At least one of the growth data of the roots of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of photographing the photographed image is stored.

More specifically, the data storage unit 106 stores the soil of a field 220 for cultivating a predetermined plant (in this example, sorghum) that can be cultivated in the second and third periods with the roots remaining after the first harvest. data, the image feature amount extracted from the first captured image of the field 220 in the first period, the growth data of the roots of the predetermined plant at the time of capturing the first captured image, and the carbon dioxide absorption amount of the roots of the predetermined plant. At least one of the image feature amount extracted from the second photographed image of the field 220 in the second term, the root growth data of the predetermined plant at the time of photographing the second photographed image, and the carbon dioxide absorption amount of the root of the predetermined plant memorize one and the other.

FIG. 6 is a data structure diagram of the soil data storage unit 160. As shown in FIG.
Soil data storage unit 160 stores, for example, the illustrated table. This table has a record for each combination of year and field ID. A record corresponds to a "row" in a table. This table also has items of acidity, drainage, fertility and viscosity, for example, for each combination of year and field ID. An item corresponds to a "row" in a table, and is also called a "column".

FIG. 7 is a data structure diagram showing the time and average color in the growth process as feature data.
The feature amount data shown in FIG. 7 is shown in a table format. The feature amount data has a record indicating the average color in each period for each combination of year and field ID. In other words, the feature amount data is set as data of year, field ID, and weekly average color in each period. That is, the average color of the field range (example of image feature amount) is set in the field of each week in each period. When the degree of greenness of the field range is used as the image feature amount, the degree of greenness of the field range is set in the field of each week in each period. FIG. 7 shows an example in which the average color is used as the image feature amount. However, if the plant height is used as the image feature amount, the plant height is set in the field of each week in each period. A field corresponds to a data value setting area and is also called a "cell".

FIG. 8 is a diagram showing an example of the data structure held by the root data storage unit 166. As shown in FIG.
That is, root data storage unit 166 stores, for example, the table shown in FIG. This table has a record for each combination of year and field ID. In addition, this table is set with year, field ID, weekly root length (an example of root growth data), and root carbon dioxide absorption amount in each period as data. When root weight is used as root growth data, root weight is set in the field of each week in each period instead of root length. When root volume is used as root growth data, root volume is set in the field of each week in each period instead of root length.

FIG. 9 is a flow chart showing the process in the test data collection phase (S10).
The soil data acquisition unit 140 acquires soil data from each field 220 (S20). The soil data acquiring unit 140 may acquire the soil data of the farm field 220 received from the farm field terminal 200 by the receiving unit 182 or the soil data of the farm field 220 accepted by the input unit 122 .

The growth management device 100 acquires satellite images and root growth data for each week in the first period of each field 220 (S22). Specifically, the image acquisition unit 142 uses the communication processing unit 108 to receive satellite images from the image providing server 300 . Satellite images are stored in the image storage unit 162 . At this time, the feature amount extraction unit 144 extracts image feature amounts from the satellite image and stores them in the feature amount data storage unit 164 . In addition, the root data acquisition unit 146 acquires root growth data for each week in the first period of each field 220 . Specifically, the root data acquisition unit 146 may acquire root growth data received by the communication processing unit 108 or may acquire root growth data received by the input unit 122 .

The processing of S24 will be described later in relation to Modification 1. In the embodiment, the processing of S24 may be omitted.

The growth management device 100 acquires satellite images and root growth data for each week in the second term of each field 220 (S26). The data acquisition process and storage process are the same as in S22.

The processing of S28 will be described later in relation to Modification 1. In the embodiment, the processing of S28 may be omitted.

The growth management device 100 acquires satellite images and root growth data for each week in the third term of each field 220 (S30). The data acquisition process and storage process are the same as in S22.

FIG. 10 is a configuration diagram of a neural network in the embodiment.
In the input layer of the neural network in the embodiment, acidity, drainage, fertility and viscosity, which are examples of soil data, and the average color of a field range, which is an example of image feature values, are input nodes. Also, a plurality of intermediate nodes are provided in the intermediate layer of the neural network. Furthermore, in the output layer of the neural network, the root length and the amount of carbon dioxide absorbed by the root, which are examples of root growth data, are set as output nodes. Only one of the root length and the amount of carbon dioxide absorbed by the root may be used as the output node. There is a full connection between the input node of the input layer and the intermediate node of the intermediate layer. Further, there is a full connection between the intermediate node of the intermediate layer and the output node of the output layer. This neural network assumes an estimation model in which the root length and the carbon dioxide absorption amount of the root are determined according to the values of the soil data and the image feature values. This neural network is used together in the first estimation model of the first period, the second estimation model of the second period, and the third estimation model of the third period.

FIG. 11 is a data structure diagram of teacher data for the first term in the embodiment.
The teacher data for the first period is prepared to generate the first estimation model for the first period. The teacher data in this example is in table format. The teacher data has records for each combination of year, field ID, period and week. In addition, each record has the items of acidity, drainage, fertility, viscosity, and average color of the field range corresponding to the input node shown in FIG. 10, and root length and root carbon dioxide absorption corresponding to the output node. has an item. One record corresponds to one sample in the teacher data. It is assumed that each sample in the teacher data can be identified according to each combination of year, field ID, period and week.

For example, the first record shown corresponds to the year 2020, field ID: A1, the combination of the first period and the first week. In the field 220a with field ID: A1 in 2020, the acidity is B1, the drainage is C1, the fertility is D1, and the viscosity is E1. These values are copied from the soil data storage section 160 by the teacher data generation section 148 . It also shows that the average color of the satellite image taken in the first week of the first term in the field 220a with field ID: A1 in 2020 was F1-1-1. This value is copied from the feature amount data storage unit 164 by the teacher data generation unit 148 . Furthermore, the root length measured in the first week of the first term in the field 220a of the field ID: A1 in 2020 is G1-1-1, and the root carbon dioxide calculated based on the root length It shows that the amount of absorption was H1-1-1. These values are copied from the root data storage unit 166 by the teacher data generation unit 148 .

Using this first term training data, the weight data that is the optimal solution is learned by the neural network. The learned weight data is stored in the estimation model storage unit 170 . The learning procedure itself using a neural network may be conventional technology.

FIG. 12 is a data structure diagram of teacher data for the second term in the embodiment.
The teacher data for the second period is prepared to generate the second estimation model for the second period. The structure of the training data for the second term is the same as the training data for the first term shown in FIG. For example, the illustrated first record corresponds to the year 2020, field ID: A1, the combination of the second period and the first week. The teacher data for the third period has a similar configuration, and values related to the third period are set.

FIG. 13 is a flow chart showing the process in the estimation model generation phase (S12).
As described above, the teacher data generation unit 148 refers to the soil data storage unit 160, the feature amount data storage unit 164, and the root data storage unit 166 to generate the first term teacher data (FIG. 11) (S40). .

The estimation model generation unit 150 inputs the teacher data for the first period and activates the learning process by the learning engine 152 (S42). The learning engine 152 executes learning processing and generates a first estimation model (S44). The first estimation model including weight data is stored in estimation model storage section 170 .

The teacher data generation unit 148 similarly generates teacher data for the second term (S46). The estimation model generation unit 150 inputs the second term teacher data and activates the learning process by the learning engine 152 (S48). The learning engine 152 executes learning processing to generate a second estimation model (S50). The second estimation model including weight data is stored in estimation model storage section 170 .

The teacher data generation unit 148 similarly generates teacher data for the third period (S52). The estimation model generation unit 150 inputs the third term teacher data and activates the learning process by the learning engine 152 (S54). The learning engine 152 executes learning processing and generates a third estimation model (S56). The third estimation model including weight data is stored in estimation model storage section 170 .

This process is performed before the estimation model usage phase. In this example, it takes place between the end of the third period in 2020 and the sowing of seeds in 2021. After that, we move to the operation stage.

FIG. 14 is a diagram showing the process in the estimation model utilization phase (S14).
The image acquisition unit 142 waits for the photographing timing (see FIG. 2) of each field 220 (S60). The image acquisition unit 142 acquires a satellite image of the field (S62). The satellite image acquisition method is the same as in the test data acquisition phase (FIG. 9). The feature amount extraction unit 144 extracts the image feature amount (S64). The extraction of the image feature quantity is also the same as in the test data collection phase (FIG. 9).

The root growth estimation unit 154 selects an estimation model according to the period at that time (S66). That is, the root growth estimation unit 154 selects the first estimation model if the imaging timing is the first period, selects the second estimation model if the imaging timing is the second period, and selects the second estimation model if the imaging timing is the third period. If so, select the third estimation model.

The estimation model utilization unit 156 applies the soil data of the field 220 and the image feature quantity extracted in S62 to the estimation model. That is, the estimated model utilization unit 156 sets these values to the input nodes of the neural network that performs calculations using the estimated model (S68). The root growth estimator 154 obtains values of root length and root carbon dioxide absorption from the output node as the calculation result of the estimation model (S70).

Then, the output unit 120 outputs (for example, displays on a display) the root length and the amount of carbon dioxide absorbed by the root (S72). The transmission unit 180 may transmit the root length and the carbon dioxide absorption amount of the root to the agricultural field terminal 200 .

In the embodiment, machine learning provides a mechanism for estimating root growth data and root carbon dioxide emissions. In particular, when the roots that grew in the first period were left behind, it was not possible to easily grasp how the roots continued to grow in the second period.

[Modification 1]
In the second term, the roots inherited from the first term will grow further. Therefore, in the second period, it may start with small roots because the first harvest was early, or it may start with large roots because the first harvest was late. Assuming that the state of the roots at the start of the second term is uniform is also not preferable for correctly estimating the growth of the roots in the second term.

The time of the first harvest is the end of the first term and the start of the second term. In Modified Example 1, the first estimation model is used to estimate the root growth data at the time of the first cutting. Then, the estimated root growth data is added to the estimation element of the root growth data in the second period as the initial value of the root state in the second period.

FIG. 15 is a configuration diagram of a neural network in modification 1. FIG.
The first estimation model of modification 1 is generated using a neural network similar to that of the embodiment. The second and third estimation models of modification 1 are generated using the neural network shown in FIG.

In the neural network (Fig. 15) in Modification 1, an input node corresponding to the root growth data or the amount of carbon dioxide absorbed by the root at the end of the first term is added. In this example, an input node for the root length at the end of the previous period is added. Intermediate nodes and output nodes remain unchanged. It is also the same as the embodiment in that it is a full connection.

At the time of the first reaping described in FIG. 2, acquisition of photographed images and measurement of roots are performed in the same manner as in the case of the photographing timing of each week. In addition, at the time of the second harvesting as well, acquisition of photographed images and measurement of roots are performed in the same manner as in the case of photographing timing of each week. Extraction of the image feature amount from the captured image is also performed in the same manner as in the case of the timing of capturing each week. The storage of data is also the same as in the case of the shooting timing of each week.

The training data for the first period in modification 1 is the same as in the case of the embodiment (FIG. 11), and the generation of the first estimation model is also the same.

FIG. 16 is a data structure diagram of teacher data for the second period in Modification 1. As shown in FIG.
In the second term teacher data in Modification 1, an item of root length at the end of the first term is added. The root length measured at the first cutting is set in the field of this item.

FIG. 17 is a data structure diagram of teacher data for the third term in Modification 1. As shown in FIG.
Similarly, in the teacher data for the third term in Modification 1, an item of root length at the end of the second term is added. The root length measured at the second cutting is set in the field of this item.

Therefore, in the process of the test data collection phase shown in FIG. 9, after the process of S22, the growth management device 100 acquires the satellite image and root growth data of each field 220 at the time of the first harvest (S24 ). After the process of S26, the growth management apparatus 100 acquires the satellite image and root growth data of each field 220 at the time of the second harvest (S28). The acquisition method of the satellite image, the method of extracting the image feature amount from the satellite image, the acquisition method of the root growth data, and the like are the same as in the case of the timing of photographing each week.

Regarding the processing in the estimation model generation phase (S12), in S46 of FIG. 13, the teacher data for the second term (FIG. 16) is generated by adding the root length at the end of the first term (at the time of the first reaping). In S48, the estimation model generation unit 150 inputs this second term teacher data and generates a second estimation model in the second modification. In S52, teacher data for the third term (FIG. 17) is generated by adding the root length at the end of the second term (at the time of the second reaping). In S54, the estimation model generation unit 150 inputs this third period teacher data and generates the third estimation model in the second modified example.

Regarding the processing in the estimation model usage phase (S14), when the first estimation model is selected in S66 of FIG. 14, the same processing as in the embodiment is performed.

In Modification 1, the receiving unit 182 receives the notification from the field terminal 200 at the time of the first reaping. Upon receiving the first harvesting notification, the image acquisition unit 142 acquires the satellite image of the field 220 in the same manner as in S62, and the feature amount extraction unit 144 extracts the image feature amount in the same manner as in S64. Then, the estimation model utilization unit 156 applies the soil data of the farm field 220 and the extracted image feature amount to the first estimation model, and the root growth estimation unit 154 calculates the root length and the root length at the time of the first harvesting. Get the value of the carbon dioxide uptake of .

In the second term after this, when the second estimation model is selected in S66, the estimation model utilization unit 156 inputs the root length at the time of the first cutting and the root length at the end of the first term in the neural network. Set to node. That is, the estimation model utilization unit 156 applies the root length at the time of the first cutting to the second estimation model in addition to the soil data and the image feature value, and the root growth estimation unit 154 applies the root length at the time of the first cutting to the second estimation model. The values of root length and root CO2 uptake in the second period are obtained considering the root length in the second period.

Also, the receiving unit 182 receives a notification from the field terminal 200 at the time of the second harvesting. Upon receiving the second harvesting notification, the image acquisition unit 142 acquires a satellite image of the field in the same manner as in S62, and the feature amount extraction unit 144 extracts the image feature amount in the same manner as in S64. Then, the estimation model using unit 156 applies the soil data of the farm field 220 and the extracted image feature amount to the second estimation model, and the root growth estimation unit 154 calculates the root length and the root length at the time of the second harvesting. Get the value of the carbon dioxide uptake of .

In the third term after this, when the third estimation model is selected in S66, the estimation model utilization unit 156 inputs the root length at the time of the second harvesting and the root length at the end of the first term in the neural network. Set to node. That is, the estimation model utilization unit 156 applies the root length at the time of the second cutting to the third estimation model in addition to the soil data and the image feature value, and the root growth estimation unit 154 applies the root length at the time of the second cutting to the third estimation model. The values of root length and root carbon dioxide uptake at the third stage are obtained considering the root length at

To summarize, the data storage unit 106 stores the soil data, the image feature amount extracted from the first captured image of the field in the first period, and the first captured image as the first-term teacher data ( FIG. 11 ). At least one of the growth data of the roots of the predetermined plant and the carbon dioxide absorption amount of the roots of the predetermined plant at the time is stored. The data storage unit 106 stores the soil data, the image feature amount extracted from the second photographed image of the field in the second term, and the predetermined Stores at least one of plant root growth data and predetermined plant root carbon dioxide absorption, and at least one of predetermined plant root growth data and predetermined plant root carbon dioxide absorption at the end of the first term. do. The data storage unit 106 stores the soil data, the image feature amount extracted from the third photographed image of the field in the third term, and the predetermined Stores at least one of plant root growth data and predetermined plant root carbon dioxide absorption, and at least one of predetermined plant root growth data and predetermined plant root carbon dioxide absorption at the end of the second period. do.

The estimation model generating unit 150 uses the neural network of FIG. 10 and the first-term teacher data (FIG. 11) to determine the roots of the predetermined plant at the time of the first photographing from the soil data and the image feature amount of the first photographed image. machine learning to generate a first estimation model for estimating at least one of the growth data of the plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant. The estimation model generation unit 150 uses the neural network of FIG. 15 and the teacher data of the second period (FIG. 16) to generate soil data, the image feature amount of the second captured image, and the number of predetermined plants at the end of the first period. estimating at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of the second photographing from at least one of the root growth data and the carbon dioxide absorption amount of the root of the predetermined plant; Generate an inference model by machine learning. Furthermore, the estimation model generation unit 150 uses the neural network of FIG. 15 and the teacher data of the third period (FIG. 17) to generate the soil data, the image feature amount of the third captured image, and the predetermined At least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the predetermined plant at the time of the third imaging is estimated from at least one of the root growth data of the plant and the carbon dioxide absorption amount of the predetermined plant root. A third estimation model is generated by machine learning.

As a premise for estimating the second period in the operation stage, the root growth estimation unit 154 uses the first estimation model to obtain the soil data and the image feature amount extracted from the first captured image at the end of the first period. At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the root of the predetermined plant at the end of the first term is estimated from the above. Then, in the second estimation, the root growth estimating unit 154 uses the second estimation model, the soil data, the image feature amount extracted from the second captured image, and the estimated end point of the first period. Based on at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the predetermined plant roots, at least the root growth data of the predetermined plant and the carbon dioxide absorption amount of the predetermined plant roots at the time of capturing the second captured image. Estimate one.

Furthermore, as a premise for estimating the third period, the root growth estimation unit 154 uses the second estimation model, from the soil data and the image feature amount extracted from the second captured image at the end of the second period, At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the root of the predetermined plant at the end of the second period is estimated. Then, in the third period estimation, the root growth estimation unit 154 uses the third estimation model to obtain the soil data, the image feature amount extracted from the third captured image, and the estimated end point of the second period. Based on at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the predetermined plant roots, at least the root growth data of the predetermined plant and the carbon dioxide absorption amount of the predetermined plant roots at the time of capturing the third captured image. Estimate one.

The state of the roots at the beginning of the second period is unknown because they are in the soil. Therefore, the first estimation model is used to estimate the state of the last root in the first period, and this is incorporated into the second estimation model. Furthermore, the state of the roots at the beginning of the third period is also unknown because it is in the soil. Therefore, the second estimation model is used to estimate the state of the last root in the second term, and this is incorporated into the third estimation model. These ideas are specific to plants that can be multi-cropped with their roots left intact, and are not related to single crops or trees. In particular, there has been no conventional focus on the roots of such plants. The motive for estimating the state of the roots in the second and third periods is due to the inventor's excellent idea, which greatly contributes to the promotion of carbon dioxide reduction business using the roots of multi-season crops (ex. sorghum). do.

[Modification 2]
In the embodiment, the growth management device 100 has the function of a machine learning device that generates an estimation model by machine learning, and the function of an estimation device that obtains root growth data and root carbon dioxide absorption from the estimation model. However, a machine learning device and an estimation device may be provided separately.

FIG. 18 is a configuration diagram of a growth management system.
The machine learning device 190 performs processing in the test data collection phase (S10) and the estimation model generation phase (S12), and generates an estimation model by machine learning. In the estimation model utilization phase (S14), the estimation device 192 acquires satellite images and uses the estimation model to obtain root growth data and root carbon dioxide absorption. Therefore, the estimation model generated by the machine learning device 190 is transferred to the estimation device 192 before moving from the estimation model generation phase (S12) to the estimation model use phase (S14). Machine learning device 190 is used, for example, by service providers who provide estimation models. The estimator 192 is used, for example, by farmers using estimation models.

[Modification 3]
The growth management apparatus 100 has a carbon price calculation unit (not shown) that calculates a carbon price (per share) by multiplying the estimated carbon dioxide absorption amount (per share) by a carbon price (rate) per unit amount. ) may be provided. The carbon price calculation unit may calculate the carbon price (per unit area) by multiplying the estimated amount of carbon dioxide absorption by the roots (per unit area) by the carbon price (rate) per unit amount. The output unit 120 may output (for example, display on a display) the carbon price (per share) or the carbon price (per unit area). The transmission unit 180 may transmit the carbon price (per share) or the carbon price (per unit area) to the field terminal 200 .

The estimation device 192 also includes a carbon price calculation unit that calculates the carbon price (per share) by multiplying the estimated carbon dioxide absorption amount (per share) by the carbon price (rate) per unit amount. You may do so. The carbon price calculation unit may calculate the carbon price (per unit area) by multiplying the estimated amount of carbon dioxide absorption by the roots (per unit area) by the carbon price (rate) per unit amount. The output unit 120 may output (for example, display on a display) the carbon price (per share) or the carbon price (per unit area). The transmission unit 180 may transmit the carbon price (per share) or the carbon price (per unit area) to the field terminal 200 .

[Other Modifications]
Instead of the satellite image, a drone image obtained by capturing the farm field 220 from above with a drone camera flying over the farm field 220 may be used. In the case of drone images, there are more pixels than satellite images, so there is an advantage that it is easy to obtain the ratio of green in the field range.

Instead of the satellite image, a camera image captured from obliquely above the field 220 or horizontally by a camera installed in the field 220 (hereinafter referred to as "installed camera") may be used. A camera image taken from the horizontal direction has the advantage that the height of the plant can be easily obtained.

In the embodiment and modification, an example of obtaining both root growth data and root carbon dioxide uptake was mainly shown, but only one of the root growth data and root carbon dioxide uptake may be used. Further, since the root growth data and the amount of carbon dioxide absorbed by the roots are strongly correlated, both may be used interchangeably as indicators in the embodiment and the modified example.

In the test stage, the field 220 may be divided into a plurality of areas, each area may be treated as one field 220, and test cultivation may be performed by changing the timing of harvesting and the soil. In this way, many samples can be included in the training data.

It should be noted that the present invention is not limited to the above embodiments and modifications, and can be embodied by modifying the constituent elements without departing from the scope of the invention. Various inventions may be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments and modifications. Also, some constituent elements may be deleted from all the constituent elements shown in the above embodiments and modifications.

Claims

Soil data of a field in which a predetermined plant is cultivated, image feature values extracted from a photographed image of the field, root growth data of the predetermined plant at the time of photographing the photographed image, and carbon dioxide of the root of the predetermined plant. a storage unit that stores at least one of the absorption amounts;
At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of photographing is estimated from the soil data and the image feature amount extracted from the photographed image. A machine learning device, comprising: a model generation unit that generates an estimation model by machine learning.
a storage unit that stores soil data of a field in which a predetermined plant is cultivated;
an image acquisition unit that acquires a photographed image of the field;
an extraction unit that extracts an image feature amount from the acquired captured image;
Soil data of a field in which the predetermined plant was cultivated in the past and an image feature amount extracted from the photographed image of the field in the past cultivation are used as input data, and the photographing time of the photographed image in the past cultivation. The stored soil data and the acquired At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the acquired captured image is estimated from the image feature amount extracted from the captured image. an estimating device, comprising: an estimating unit;
Soil data of a field for cultivating a predetermined plant that can be cultivated in the second period with the roots remaining after the first harvesting period, and an image feature amount extracted from the first photographed image of the field in the first period. from at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the first photographing of the first photographed image, and the second photographed image of the field in the second period; a storage unit that stores the extracted image feature amount, and at least one of root growth data of the predetermined plant at the time of the second photographing of the second photographed image and carbon dioxide absorption amount of the root of the predetermined plant;
At least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant at the time of the first photographing is estimated from the soil data and the image feature amount of the first photographed image. Generate a first estimation model by machine learning,
At least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant at the time of the second photographing is estimated from the soil data and the image feature amount of the second photographed image. A machine learning device comprising: a model generation unit that generates a second estimation model by machine learning.
a storage unit for storing soil data of a field for cultivating a predetermined plant that can be cultivated for a second period with roots remaining after the first harvest;
an image acquisition unit that acquires a first photographed image of the farm field in the first period or a second photographed image of the farm field in the second period;
an extraction unit that extracts an image feature amount from the first captured image or the second captured image;
When the first photographed image is acquired, soil data of a field in which the predetermined plant was cultivated in the past cultivation, and an image feature amount extracted from the photographed image of the field in the first period in the past cultivation. Machine learning using as input data and at least one of root growth data of the predetermined plant at the time of photographing the photographed image in the first period in the past cultivation and at least one of carbon dioxide absorption amount of the root of the predetermined plant as output data. Using the obtained first estimation model, from the stored soil data and the image feature amount extracted from the first photographed image, the root shape of the predetermined plant at the time of photographing the first photographed image. estimating at least one of growth data and the amount of carbon dioxide absorbed by the roots of the predetermined plant;
When the second photographed image is acquired, soil data of the field in which the predetermined plant was cultivated in the past cultivation, and an image feature amount extracted from the photographed image of the field in the second period in the past cultivation. is input data, and at least one of the root growth data of the predetermined plant at the time of capturing the captured image in the second period in the past cultivation and the carbon dioxide absorption amount of the root of the predetermined plant is used as output data Using the learned second estimation model, the root of the predetermined plant at the time of capturing the second captured image from the stored soil data and the image feature amount extracted from the second captured image. and an estimating unit that estimates at least one of the growth data of the plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant.
Soil data of a field for cultivating a predetermined plant that can be cultivated in the second period with the roots remaining after the first harvesting period, and an image feature amount extracted from the first photographed image of the field in the first period. from at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the first photographing of the first photographed image, and the second photographed image of the field in the second period; at least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of the second photographing of the second photographed image; and the end of the first term. a storage unit that stores at least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant;
At least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant at the time of the first photographing is estimated from the soil data and the image feature amount of the first photographed image. Generate a first estimation model by machine learning,
From the soil data, the image feature amount of the second photographed image, and at least one of the root growth data of the predetermined plant at the end of the first period and the carbon dioxide absorption amount of the root of the predetermined plant, a model generating unit that generates, by machine learning, a second estimation model that estimates at least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant at the time of the second photographing. learning device.
a storage unit for storing soil data of a field for cultivating a predetermined plant that can be cultivated for a second period with roots remaining after the first harvest;
an image acquisition unit that acquires a first photographed image of the farm field at the end of the first period and further acquires a second photographed image of the farm field in the second period;
an extraction unit that extracts an image feature amount from the first captured image and further extracts an image feature amount from the second captured image;
Soil data of a field in which the predetermined plant was cultivated in the past cultivation and an image feature amount extracted from a photographed image of the field in the first period in the past cultivation are used as input data, and the 1 in the past cultivation Using a first estimation model machine-learned using at least one of the root growth data of the predetermined plant and the carbon dioxide absorption amount of the root of the predetermined plant at the time of photographing the photographed image of the period as output data, Root growth data of the predetermined plant and carbon dioxide absorption by the roots of the predetermined plant at the end of the first term based on the stored soil data and the image feature amount extracted from the first photographed image. Estimate at least one of the quantities,
The soil data of the field where the predetermined plant was cultivated in the past cultivation, the image feature amount extracted from the photographed image of the field in the second period in the past cultivation, and the first period in the past cultivation. At least one of the root growth data of the predetermined plant at the end point and the carbon dioxide absorption amount of the root of the predetermined plant is used as input data, and the photographed image of the second term in the past cultivation is taken. Using a second estimation model machine-learned using at least one of root growth data of a predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant as output data, the stored soil data and the second from the image feature quantity extracted from the photographed image, and at least one of the growth data of the roots of the predetermined plant estimated at the end of the first period and the carbon dioxide absorption amount of the roots of the predetermined plant and an estimating unit that estimates at least one of root growth data of the predetermined plant and carbon dioxide absorption amount of the root of the predetermined plant at the time of capturing the second captured image.
Soil data of a field in which a predetermined plant is cultivated, image feature values extracted from a photographed image of the field, root growth data of the predetermined plant at the time of photographing the photographed image, and carbon dioxide of the root of the predetermined plant. a function of storing at least one of the absorbed amounts;
At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of photographing is estimated from the soil data and the image feature amount extracted from the photographed image. A program that makes a computer demonstrate the function of generating an inference model by machine learning.
a function of storing soil data of a field in which a predetermined plant is cultivated;
a function of acquiring a photographed image of the field;
A function of extracting an image feature amount from the acquired captured image;
Soil data of a field in which the predetermined plant was cultivated in the past and an image feature amount extracted from the photographed image of the field in the past cultivation are used as input data, and the photographing time of the photographed image in the past cultivation. The stored soil data and the acquired At least one of the root growth data of the predetermined plant and the amount of carbon dioxide absorbed by the roots of the predetermined plant at the time of capturing the acquired captured image is estimated from the image feature amount extracted from the captured image. A program that makes a computer demonstrate functions.
The estimation device according to claim 2, 4, or 6, further comprising a carbon price calculation unit that calculates a carbon price based on the estimated carbon dioxide absorption amount.