JP6971453B2 - Seedling data generation system, seedling discrimination system, seedling data generation program, seedling discrimination program, seedling data generation device, seedling discrimination device - Google Patents

Description

The present invention is a seedling data generation system, a seedling discrimination system, a seedling data generation program, a seedling discrimination program, a seedling data generation device, and a seedling discrimination for selecting (discriminating) the quality of the growth of the original seedlings of the grafted plant and the seedlings after grafting. Regarding the device.

In recent years, the low food self-sufficiency rate has become a major problem in Japan. And if global warming, which is a problem all over the world, progresses further, it is feared that it will further hurt Japan's agriculture and accelerate the low food self-sufficiency rate. Although Japan is also providing assistance to Japanese farmers through all measures, it has an explosive effect on Japan's food self-sufficiency rate due to the aging of farmers and the shortage of successors. Not.

Farmers purchase pre-raised plant seedlings to improve production efficiency amid climate change due to global warming, the aging population, and a shortage of successors. By purchasing and cultivating raised seedlings, farmers can focus on plant production and work efficiency in farmers is improved.
Farmers purchase seedlings they grow from seedling producers and distributors. It is very important for seedling producers and distributors to maintain the quality of seedlings, because the quality of purchased seedlings greatly affects the cultivation status of seedlings by farmers and the production and sales of plants.

Seedling producers and distributors provide and sell seedlings to farmers after raising seedlings to some extent, but grafting is generally used as a method for raising seedlings. This is because grafting can avoid damage from pests and increase the number of harvests, and is known for its stable quality.
Grafting is the process of adhering two or more plants with an artificially cut cut surface to form a single individual. The seedlings that have been grafted are called grafted seedlings, and the seedlings that are the source of the grafts are called original seedlings. The upper plant of the grafted seedling is called a scion, and the lower plant is called a rootstock. Grafting is often used for vegetables such as tomatoes, eggplants, cucumbers and watermelons, and fruits.

Grafted seedlings are, for example, shipped to farmers through the following steps at a seedling producer / distributor.
1. 1. Order: Make a contract such as "How many seeds are sown in a plug tray (resin container) with how many holes" from a farmer who is a customer.
2. 2. Sowing: Sowing (sowing) is performed on the plug tray using a vacuum sowing machine.
3. 3. Germination management: Seeds sown in plug trays are kept warm and moist in the germination cabinet. When the seedlings sprout in the soil, they are moved to the initial nursery building to manage germination.
4. Initial seedling raising: After germination, seedlings are raised before grafting. In particular, plant height, number of leaves, and stem thickness are controlled.
5. Grafting: Grafting by combining scion and rootstock. A tube or clip may be attached to the grafted part.
6. Curing acclimation: Grafted seedlings are sent to the curing room to settle the seams (cut surfaces).
7. Plug shipment: After grafting, seedlings with a relatively small number of leaves that have undergone curing acclimation are shipped in a plug tray without being potted.
8. Potting: If not shipped in the plug tray of step 7, the grafted seedlings are moved to the pot one by one.
9. Secondary seedling raising: Grafted seedlings are moved from the plug tray to the pot and further seedlings are raised.
10. Sorting: After curing and secondary raising of seedlings, the workers sort the grafted seedlings with the quality suitable for shipping.
11. shipping

In order for seedling producers and distributors to graft trees, they need well-grown original seedlings that are not immature or malformed. Moreover, even after grafting, the quality that can withstand shipping cannot be maintained unless the seedlings are well-grown. In this way, it is very important for seedling producers and distributors to determine the quality of the growth of original seedlings and grafted seedlings for each grafting process.
However, at seedling producers and distributors, original seedlings and grafted seedlings were visually managed and sorted based on the many years of experience of workers, especially skilled workers. In addition, the quality of management and selection by skilled workers varied depending on the fatigue and physical condition of the person. Moreover, the know-how of skilled workers was not educated by inexperienced people.

In addition, indicators that determine that the growth of original seedlings and grafted seedlings is good, such as the site of the seedling (stem diameter of each epicotyl and hypocotyl, hypocotyl length, etc.) and its numerical range (area, length, color) Etc.) vary greatly depending on the individual expert. Even if the part that is the index of discrimination is the same for each expert, it is only a sensuous judgment such as "If this part is about this length, it can be judged that training is good", and training is possible. In many cases, it has not been reduced to a specific numerical range that can be determined to be good.
Furthermore, when the variety and process of seedlings are diverse, it is necessary to memorize the index for determining that the cultivation is good for each individual variety and process, which is a burden on the worker.

For this reason, seedling producers / distributors have developed devices, methods, systems, etc. that automatically determine the quality of original seedlings and grafted seedlings. For example, a method of imaging the original seedlings in the initial seedling raising process, automatically extracting immature seedlings and malformed seedlings by analyzing the captured data of the captured two-dimensional image, and selecting seedlings that do not meet the shipping standards. Is known (Japanese Patent Laid-Open No. 09-023747).

In Japanese Unexamined Patent Publication No. 09-023747, attention is paid to the fact that immature seedlings have a short plant height and a low center of gravity. The height of the center of gravity of the original seedling is measured. If the height of the center of gravity of the original seedling is low, it is determined to be an immature seedling.
If the height of the center of gravity of the original seedling is not low, measure the area of the cotyledon while rotating it around the cup axis, detect the position where the area of the cotyledon is the maximum area, and align the original seedling at that position. ing. At the aligned position, the inclination of the original seedling from the vertical direction is measured, and if the inclination angle is more than a certain level, it is determined to be a malformed seedling (a seedling bent from the cup axis). Furthermore, the original seedlings are imaged from above with a camera, and the area of the cotyledons is measured by analyzing the captured image data. If it is small, etc.).

On the other hand, in recent years, the use of machine learning by artificial intelligence has been attracting attention in all fields. For example, a system for discriminating the material of waste using artificial intelligence is known (Japanese Unexamined Patent Publication No. 2017-109197).
In JP-A-2017-109197, the waste is imaged to obtain image data, and the weight and volume of the waste are measured to calculate the specific gravity. Then, the worker creates the teacher data by confirming the image data, the weight, the volume, and the specific gravity, identifying and inputting the material of the waste (wood, plastic, iron, non-iron, etc.).

Further, it is desirable to prepare a large amount of teacher data necessary for machine learning, but it is difficult to actually prepare a large amount of teacher data. Therefore, a model that improves the accuracy of discrimination with a small amount of teacher data is known (for example, Japanese Patent Application Laid-Open No. 2010-287065).
Japanese Patent Application Laid-Open No. 2010-287065 uses a collation image generation probability model and a fluctuation probability model in biometric authentication based on tree-like biometric information such as blood vessels to improve the accuracy of biometric authentication with a small amount of teacher data. ..

Japanese Unexamined Patent Publication No. 09-023747 Japanese Unexamined Patent Publication No. 2017-109197 Japanese Unexamined Patent Publication No. 2010-287065

According to JP-A-09-023747, immature seedlings and malformed seedlings can be automatically discriminated, and by removing these seedlings, high-quality original seedlings can be selected. Then, if a high-quality original seedling is grafted at a seedling producer / distributor, a high-quality grafted seedling can be produced.
However, it is difficult to automatically determine the original seedlings when the original seedlings are only in the state of cotyledons, that is, the state of only the first short period of the initial seedlings, and then the original seedlings grow and move to another process. ..
Further, the imaging is mainly performed from the side surface direction of the original seedling, and in some cases, the imaging from the upper direction is only additionally performed. Furthermore, these images are represented by two-dimensional images. Therefore, even if the original seedlings are well-bred, if the cotyledons are slightly bent, the two-dimensional image may not be able to accurately grasp the appearance of the original seedlings (the appearance and shape of the entire seedlings). If it is not possible to accurately grasp the appearance of the original seedlings, high-quality original seedlings will be discarded, which may lead to loss for the seedling producers and distributors.
Of course, not only the original seedlings but also the grafted seedlings can lead to loss if the seedlings cannot be accurately grasped.

According to JP-A-2017-109197, when a worker prepares a large amount of teacher data, artificial intelligence learns the material of waste and automatically identifies the material to determine whether it is waste or not. can do.
Similarly, if a seedling producer / distributor develops a seedling quality determination system using artificial intelligence, it is considered that the quality of seedlings can be automatically determined without relying on an expert. However, the seedlings of plants continue to grow day by day, and the appearance of the seedlings is changing from moment to moment. The varieties of seedlings to be grafted and their processes are diverse depending on the combination, and the index for discriminating seedlings by a skilled person is sensuous. Therefore, it is difficult for a skilled person to prepare a large amount of teacher data in all varieties and processes of seedlings.

According to Japanese Patent Application Laid-Open No. 2010-287065, a fluctuation probability model based on a probabilistic index is adopted for personal authentication (biometric authentication) by artificial intelligence. By applying the fluctuation probability model to the field of biometric authentication, it is possible to improve the accuracy of personal authentication even for biometric information having large individual differences such as veins in the palm of the hand.
However, in order to improve the accuracy of personal authentication, some teacher data is required.

It is also conceivable to regard the sensory index of a skilled person in the discrimination of seedlings as a "stochastic index" and apply the fluctuation probability model exemplified in JP-A-2010-287065 to the index for discrimination of seedlings. However, there are more varieties and processes of seedlings than humans (living bodies), and individual differences are large. Furthermore, there are various requests from farmers who are customers, and it is not appropriate to apply a fluctuation probability model that only determines "whether or not the person is the person" to the determination of seedlings.
In addition, since there are many varieties and processes for seedlings and individual differences are large, it is considered difficult to prepare a large amount of teacher data for each variety and process.

An object of the present invention is to provide a seedling data generation system, a seedling data generation program, and a seedling data generation device that accurately grasp the seedling appearance of seedlings and do not require a large amount of teacher data.
Another object of the present invention is to provide a seedling discrimination system, a seedling discrimination program, and a seedling discrimination device that accurately grasps the seedling appearance of seedlings and automatically sorts (discriminates) high-quality seedlings.

In the present invention, geometric transformation is applied to seedlings. First, one image data representing the whole image of the seedlings determined to be good for growing for each variety and process is acquired. Then, an index for determining that the seedlings are well grown is set in advance as a parameter, and a plurality of image data for which it is determined that the seedlings are well grown are automatically generated from one image data.
That is, according to the present invention according to claims 1, 4 , and 6, among the image data showing the whole image of the seedlings to be raised, the variety name of the seedlings, and the plurality of steps until the raising of the seedlings is completed. The seedling basic data acquisition means for acquiring seedling basic data including the name of one step of the above, and parameters including at least one part of the seedling and the numerical range of the part are acquired by the seedling basic data acquisition means. For the seedling basic data of one seedling which is provided with a seedling memory data generation means for generating seedling memory data in association with the image data of the seedling basic data and which is determined to be good for growing for each variety and process. When the part of the seedling and the numerical range of the part, which are indicators for determining that the growth is good, are set as parameters, the seedling storage data generation means is the image acquired by the seedling storage data generation means. A plurality of image data of seedlings according to the above parameters are generated from the data and added to the image data of the basic seedling data, and the seedling storage data is generated from the basic seedling data.

Further, in the present invention, image data showing the whole image of seedlings is acquired for each variety and process. Teacher data is generated and stored by associating the image data of seedlings, varieties, processes, and the results of determining whether the seedlings are good or bad. Then, the quality of seedling growth is automatically determined by machine learning from the generated teacher data and seedling image data.
That is, according to the present invention according to claims 3, 5 and 7, an imaging means for capturing image data representing an overall image of seedlings to be raised, a variety name of the seedlings, and until the growth of the seedlings is completed. The image pickup is provided with an input means for inputting the name of one of a plurality of steps, a control means for determining whether or not the seedlings to be raised are good or not, and a storage means for storing data. The means captures image data including three-dimensional data of one seedling determined to be good for growing for each seedling variety and process, and the input means further serves as an index for determining that the seedling is good to grow. The site and the numerical range of the site are input as parameters, and the control means includes an image processing means for generating a plurality of image data of seedlings according to the parameter from the image data captured by the image pickup means, and the image processing means. The seedling basic data is obtained by associating the image data generated by the image processing means with the variety name input by the input means, the name of one process, and the determination result that the breeding is good. It has a teacher data generation means for adding and storing data as teacher data to the storage means, and the control means performs learning using the seedling basic data as teacher data for discrimination, and further, the image processing means. It has a machine learning means for determining whether or not seedlings are well grown from the image data generated in .

In the present invention according to claims 1, 4 , and 6, it is possible to accurately grasp the appearance of the seedlings to be raised by capturing and acquiring image data representing the entire image of the seedlings to be raised. In addition, one seedling that is determined to be well grown is selected in advance, and one of these image data is acquired. Then, by simply setting the part of the seedling as an index for determining that the seedling is good to grow and the numerical range of the part as parameters, it is possible to automatically generate a plurality of image data of the seedling with good growth by geometric transformation. can. This makes it possible to automatically and easily prepare teacher data for determining whether or not the seedlings are well grown (good or bad seedlings). Further, by using the geometric transformation instead of the fluctuation probability model, the operator can visually recognize the image data and operate the parameters.
Also in the present invention according to claims 3, 5 and 7, it is possible to accurately grasp the seedling appearance of the original seedling by capturing and acquiring image data representing the whole image of the seedling to be raised. Then, by accumulating the acquired image data as teacher data and repeating learning by artificial intelligence, it is possible to automatically determine from the image data whether or not the seedlings are well grown.

The schematic diagram which shows the structure of the seedling data generation apparatus and the seedling discrimination apparatus which concerns on one Example of this invention is shown. A block diagram showing a configuration example of a seedling data generation device and a seedling discrimination device according to an embodiment of the present invention is shown. (A) shows a flow chart explaining the flow of generation of seedling memory data, and (B) shows a flow chart explaining the flow of generation of seedling basic data. An explanatory diagram for explaining the imaging of seedlings is shown. (A) to (C) show explanatory views showing an example of a captured image and a processed image, respectively. An explanatory diagram showing image data of new seedlings is shown. (A) shows a flow chart explaining the flow of quality determination of seedling growth, and (B) shows a flow chart explaining the flow of acquiring basic seedling data and image processing. The flow chart explaining the flow of the generation of the seedling memory data and the quality judgment of the seedling growth is shown.

Seedling basics to acquire basic seedling data including image data showing the overall image of the seedlings to be raised, the cultivar name of the seedling, and the name of one of a plurality of steps until the seedling growth is completed. Seedling storage data by associating a parameter including a data acquisition means, at least one part of the seedling, and a numerical range of the part with the image data of the seedling basic data acquired by the seedling basic data acquisition means. With respect to the seedling basic data of one seedling that is determined to be good for growing for each variety and process, the site of the seedling that is determined to be good for growing and the relevant seedling When the numerical range of the site is set as a parameter, the seedling storage data generation means generates a plurality of image data of seedlings corresponding to the parameters from the image data acquired by the seedling storage data generation means and seedlings. Seedling storage data is generated from the seedling basic data in addition to the image data of the basic data.

1. 1. Configuration of Seedling Data Generation System and Seedling Discrimination System Hereinafter, examples of the present invention will be described in detail with reference to the drawings. FIG. 1 shows a schematic diagram showing the configuration of a seedling data generation device and a seedling discrimination device according to an embodiment of the present invention.
The seedling data generation system 1 and the seedling discrimination system 2 of the present invention include, for example, a seedling data generation device 10-1, a seedling discrimination device 10-2, an external database 20, an input means 31 and a display means 32. The terminal 30 and the image pickup means 40 are connected to each other by a network 50 so as to be communicable with each other. The terminal 30 is operated by an operator.
In addition, in FIG. 1, for the sake of simplification of the description, it is described that the seedling data generation device 10-1 and the seedling discrimination device 10-2 are incorporated into one device, but the present invention is not limited to this. The seedling data generation device 10-1 and the seedling discrimination device 10-2 may be incorporated into one device or may be incorporated into different devices.
Further, the configuration by the database 20, the terminal 30, the imaging means 40, and the network 50 is an example for making the seedling data generation system 1 and the seedling discrimination system 2 function, and is not limited thereto. For example, depending on the environment, it may not be necessary to connect via the network 50.

The seedling data generation device 10-1 and the seedling discrimination device 10-2 for the seedling data generation system 1, the seedling discrimination system 2 are, for example, server-type information processing devices, and information to the terminal is received in response to a request from the terminal 30. Perform operations such as transmission. Inside the terminal 30, an electronic component such as a CPU (Central Processing Unit) having an information processing function and a memory is provided.
The seedling data generation device 10-1 and the seedling discrimination device 10-2 are not limited to the server-type information processing device.

The database 20 has a master and stores each information in the master. It should be noted that each information (data) does not have to be in a table format data structure, and may be in a format other than a table such as a list or a queue.

The terminal 30 is an information processing device such as a personal computer, and includes internal electronic components such as a CPU and a flash memory for processing information. For example, an input means 31 such as a keyboard and a mouse, a display means 32 such as a monitor, and the like are connected to the terminal 30 as specific devices.
The input means 31 is a well-known device for operating (inputting) from the outside of the control means 11 (FIG. 2) in the seedling data generation device 10-1 and the seedling discrimination device 10-2 via the terminal 30. Specifically, a keyboard, a mouse, and the like correspond to the input means 31. The input result to the input means 31 is temporarily output to the control means 11 via the terminal 30.
The display means 32 displays various images via the terminal 30 based on the signals output by the control means 11 in the seedling data generation device 10-1 and the seedling discrimination device 10-2.

The image pickup means 40 is a device for acquiring image data representing the whole image of the seedlings to be grown. Examples of the image pickup means 40 include, but are not limited to, a plurality of cameras.
Further, as a method of generating image data (model; for example, a three-dimensional image) from a multi-viewpoint image captured by a plurality of imaging means 40, a visual volume crossing method can be specifically mentioned. The visual volume crossing method is image data (model; three-dimensional image) using a binary image (black and white binary image) called a silhouette image in which the silhouette of the subject is shown in black and the other areas are shown in white in each camera image. Is a method of generating.
Of course, the method for generating image data in the present invention is not limited to the visual volume crossing method. Any image data format may be used as long as it expresses image data representing the entire image of the seedlings to be grown, and specifically, it may or may not be a three-dimensional image.

The network 50 is a communication network, and is, for example, a wired or wireless communication network such as an intranet or a LAN (Local Area Network).

2. 2. Configuration of Seedling Data Generation Device and Seedling Discrimination Device FIG. 2 shows a block diagram showing a configuration example of a seedling data generation device and a seedling discrimination device according to an embodiment of the present invention.
The seedling data generation device 10-1 and the seedling discrimination device 10-2 are composed of a CPU (processor) and the like to control each part, and are also composed of a control means 11 for executing various programs and a storage medium such as a flash memory. The storage means 12 for storing information and the communication means 13 for communicating with the outside via the network 50 are provided.
The present invention is, for example, executed by the control means 11 in the server-type information analyzer, but is not limited to this, and may be executed by software executed on a general-purpose computer, or dedicated hardware. Or, it may be executed by a combination of software and hardware.

The control means 11 executes the seedling data generation program 121 and the seedling discrimination program 122 stored in the storage means 12, which will be described later, to execute the seedling basic data acquisition means 110, the image processing means 111, the parameter setting means 112, and the seedling storage data. It functions as a generation means 113, a teacher data generation means 114, and a machine learning means 115.

The storage means 12 is composed of, for example, a magnetic storage medium, an optical storage medium, a non-volatile semiconductor memory device, or the like, and stores system programs 120, 121, and system data 122 to 124. The system program is, for example, a seedling data generation program 120 related to the seedling data generation system 1 and a seedling discrimination program 121 related to the seedling discrimination system 2, and is a program read by the control means 11 to execute various processes.
The system data 122 to 124 are, for example, parameter data 122 described later, seedling storage data 123 (particularly related to the seedling data generation system 1), teacher data 124 (particularly related to the seedling discrimination system 2), and the like, and the system program 120. Data that is referenced or stored with the execution of 121.

3. 3. Seedling data generation system 3.1 Processing procedure of seedling data generation system First, the generation of seedling storage data executed by the seedling data generation system 1 will be described. FIG. 3A shows a flow chart for explaining the flow of generation of seedling storage data, and FIG. 3B shows a flow chart for explaining the flow of generation of seedling basic data. As shown in FIG. 3, the seedling data generation system 1 includes a seedling basic data acquisition step, a parameter setting step, and a seedling memory data generation step in order until the seedling storage data is generated.

As can be seen from FIG. 3 (A) together with FIG. 2, in the seedling basic data acquisition step (step S11), the seedling basic data acquisition means 110 is an image data representing the whole image of the seedling to be raised, and the seedling. Acquire basic seedling data including one of a plurality of steps until the breeding name and seedling growth are completed. Further, in the seedling basic data acquisition step (step S11), the image processing means 111 may perform image processing for processing the image data.
In the parameter setting step (step S12), a parameter (parameter data 122) including at least one part of the seedling and a numerical range of the part is set and stored by the operator via the parameter setting means 112.
In the seedling storage data generation step (step S13), the seedling storage data generation means 113 associates with the image data of the seedling basic data to generate and store the seedling storage data 123.

FIG. 4 shows an explanatory diagram illustrating the imaging of seedlings. Note that FIG. 4 has colors.
5 (A) to 5 (C) show explanatory views showing an example of a captured image and a processed image. In FIGS. 5A to 5C, the image data is represented as a two-dimensional image for ease of understanding. However, the image data representing the whole image of the seedling is sufficient, and the image data is not limited to the two-dimensional image. For example, the image data of the seedling may be a three-dimensional image. Further, in FIGS. 5A to 5C, a frame line is described to indicate that the image is captured, but the description of the frame line is not essential.
The flow of generation of seedling storage data (steps S11 to S13) will be described with reference to FIGS. 3 (B) to 5 (C).
First, in step S11, only one seedling (hereinafter, appropriately referred to as “seedling to be raised” or simply “seedling”) 60 determined to be “good to grow” for each variety and process is selected by the worker (hereinafter, referred to as “seedling to be raised” or simply “seedling”) 60. Step S11-1).
As the seedling 60 to be imaged by the seedling data generation system 1, the seedling determined to be poorly grown is not selected. Further, even if it is determined that the seedlings are well grown, only one seedling to be imaged is sufficient, and a plurality of seedlings are not required.

Varieties refer to the types of vegetables and fruits used for grafting. Vegetables and fruits, which are usually used as the source seedlings for grafting, are prepared in multiple varieties to meet customer requirements. However, if only one variety is prepared for one type of vegetable (for example, tomato), the expression of "variety" in the seedling data generation system 1 is the vegetable name (tomato, cucumber, eggplant, etc.). May be only. Needless to say, inside the system, the product type and process are usually represented by some kind of code.
The process refers to the process of grafting. In the grafting process, for example, as described above, ordering, sowing, germination management, initial seedling raising, grafting, curing acclimation, plug shipping, potting, secondary seedling raising, sorting, shipping, surplus management, surplus soil treatment / hilling production Etc., but not limited to these.
The seedling 60 to be imaged, which is shown in FIG. 4, is a cultivar called "Reiyo" of "tomato" in the process of "initial raising seedlings". That is, the cultivar name "Reiyo" and the process name "initial seedling raising" of "tomato" are input in advance by the operator via the input means 31 (step S11-2).

Then, the seedlings 60 to be raised are imaged with colors by the imaging means 40 as an image showing the whole image of the seedlings (step S11-3). Reference numeral 70 in FIG. 5A indicates image data obtained by capturing an image of the seedling 60 to be raised. The image data 70 is represented as a two-dimensional image for ease of understanding.
As long as the image pickup means 40 can capture and acquire the entire image of the seedlings 60 to be raised, that is, information representing all directions of the seedling appearance (for example, a three-dimensional image), a plurality of the image pickup means 40 may be used as shown in FIG. It may be a stand.

Then, the cultivar name (“Reiyo” of “tomato”), the process name (“initial seedling raising”), and the image data 70 of the seedlings imaged by the imaging means 40 are linked to form basic seedling data (step). S11-4). In the embodiment, the cultivar name and the process name are input in step S11-2, and then the seedlings are imaged in step S11-3, but the order of steps S11-2 and 3 may be reversed. ..
The image data 70 of the seedlings imaged by the image pickup means 40 is sent from the image pickup means to the control means 11, and the color is removed by the image processing means 111 and processed so as to be expressed only in black and white (step S11). -5, FIG. 2, FIG. 5 (B)). Reference numeral 71 in FIG. 5B indicates image data represented in black and white.

The image data 71 of the seedlings represented in black and white is further processed by the image processing means 111 so as to be represented only by the contour line (step S11-6, FIG. 5C). Reference numeral 72 in FIG. 5C indicates image data represented only by contour lines.
Then, when the image processing of steps S11-5 and 6 is performed, the image data 72 represented only by the contour line generated in step S11-6 is the image of the seedling basic data generated in step 11-4. The data is overwritten (step S11-7).

For example, the image data 72 represented only by the contour line is sent to the display means 32 and displayed on the display means together with the product type name and the process name. Then, on the image data 72 of the seedlings by the worker (mainly an expert), the part of the seedling which is an index for determining that the growth is good and the numerical range of the part (the numerical range where the growth is determined to be good). Is specified as a parameter. The designated parameter is set and stored as parameter data 122 by the parameter setting means 112 (step S12).
The parameter data 122 may be set and stored after displaying the image data 72 of the seedlings on the display means 32, or may be set and stored in the storage means 12 in advance.

As shown in FIG. 5 (C), in the imaged seedling 60 (process "initial raising seedling", "Reiyo" of the variety "tomato"), for example, the area (size) of the first true leaf 61 is "growth". It is regarded as one of the "indexes judged to be good". In the process "initial raising seedlings" and "Reiyo" of the variety "tomato", the seedling part (first true leaf) and the numerical range (area) of the part are specified as parameters. For example, in the image data 72 of the seedling, the image processing means 111 recognizes the part (range) of the seedling designated as the range of the coordinate values in the space.

FIG. 6 shows an explanatory diagram showing image data of new seedlings.
When the parameter is specified from the image data 72 of the seedling, the new part 61-1 to which the designated part 61 (first true leaf) is changed within the specified numerical range by the image processing means 111 in the control means. Multiple 61-6 are created. Then, by combining the created new parts 61-1 to 61-6 with image data other than the specified part, a new image of the seedling represented only by the contour line corresponding to the specified parameter is obtained. Data (hereinafter, simply referred to as “new seedling image data”) 72-1 to 72-6 are generated (step S13).

Then, the seedling storage data generation means 113 adds the generated new seedling image data 72-1 to 72-6 to the image data of the seedling basic data acquired in step S11, respectively, from the seedling basic data. Seedling storage data 123 is generated. The generated seedling storage data 123 is stored in the storage means 12 (step S13).

In FIGS. 6A to 6F, six new seedling image data 72-1 to 72-6 are generated from the image data 72 of one seedling, but the number is not limited to this. The number of new seedling image data to be generated from the image data 72 of one seedling within the numerical range of the specified parameter is determined, for example, by the performance of the artificial intelligence used and the capacity of the storage means 12. In addition, the worker may set a new restriction on the image data of the seedlings.
Further, in the embodiment, the parameter was changed by designating the first true leaf 61, but the designated site may be a site that serves as an "index for determining good growth" and is not limited to the first true leaf. Needless to say. Of course, the designated part is not limited to one place, and a plurality of parts may be specified and parameters may be set for each.

Further, in the embodiment, in step S13, the size (numerical range; parameter) of the first true leaf 61 was changed to generate new parts 61-1 to 61-6, but the parameter to be specified is limited to the size. Not done. For example, an angle may be specified as a parameter to rotate the specified portion, or both a size and an angle may be specified to rotate while changing the size of the specified portion. For example, if the size of the specified part is set to 1, and the size is changed (enlarged / reduced) in 0.1 increments from 0.9 to 1.1, and the rotation is specified in 90 degree increments, a new one is specified. Multiple image data of seedlings are generated.

3.2 Effect of seedling data generation system According to the seedling data generation system 1, it is possible to accurately grasp the whole image (seedling appearance) of the seedlings to be raised.
Then, according to the seedling data generation system 1, it is determined that the growth is good only by specifying the parameter (the part of the seedling and the numerical range of the part) that is an index for determining that the growth is good for each variety and process. A plurality of new seedling image data 72-1 to 72-6 can be generated from the image data 70 of one seedling. That is, a plurality of image data are automatically generated from one image data that is determined to be good for breeding. Therefore, it can be automatically and easily prepared as a plurality of image data and teacher data for determining whether or not the seedlings are good or not (good or bad of the seedlings).
Further, the image data having colors is first processed by the image processing means 111 so as to be represented only in black and white and then represented only by contours. If the image data having colors is stored in the storage means 12 as it is, the data capacity of the storage means is squeezed, but by performing image processing using only black and white and contours, only the data necessary for determining the quality of training can be obtained. It is possible to secure the data capacity of the storage means while suppressing it.
Further, by using the geometric transformation instead of the fluctuation probability model, the operator can operate the parameters while visually recognizing the image data on the display means 32.

4. About seedling discrimination system
4.1 Processing procedure of seedling discrimination system (when seedling memory data is not used as teacher data)
Next, the determination of seedling growth (determination of whether or not the seedlings are good or not) executed by the seedling discrimination system 2 will be described. FIG. 7A shows a flow chart for explaining the flow of determining the quality of seedling growth, and FIG. 7B shows a flow chart for explaining the flow of acquiring basic seedling data and image processing. As shown in FIG. 7A, the seedling discrimination system 2 includes a seedling basic data acquisition step and a machine learning step in order until the seedling growth is discriminated.
In the seedling discrimination system 2, the components common to the seedling data generation system 1 are designated by the same reference numerals and the description thereof will be omitted, and the configuration of the seedling discrimination system different from the seedling data generation system will be mainly described. FIGS. 1, 2, 4 to 6 are adopted as they are in the seedling discrimination system 2.

In the seedling discrimination system 2, the teacher data when the seedling storage data generated by the seedling data generation system 1 is "good to grow" may or may not be used. In the following description, first, a flow in which the seedling storage data generated by the seedling data generation system 1 is not used as teacher data will be described.
When the seedling storage data generated by the seedling data generation system 1 is not used as the teacher data, the seedling discrimination system 2 itself is characterized in that it acquires the seedling basic data and performs black-and-white and contour image processing of the image data (. Steps S21, S22, FIG. 7).

In the seedling basic data acquisition step (step S21), the seedling basic data acquisition means 110 is among a plurality of steps until the image data representing the whole image of the seedling to be raised, the seedling variety name, and the seedling growth are completed. Obtain basic seedling data including one step and link them. The seedling basic data becomes teacher data in the next machine learning step (step S22). Further, in the seedling basic data acquisition step (step S21), the image processing means 111 performs image processing for processing the image data.
In the machine learning step (step S22), the machine learning means 115 performs supervised learning based on the teacher data in step S22, records the learning result data as teacher data, and creates a database.

Each process will be described in detail. First, a flow in which the seedling storage data generated by the seedling data generation system 1 is not used as teacher data will be described.
The flow of acquisition of seedling basic data and image processing (step S21) will be described with reference to FIG. 7B.
First, the worker selects seedlings 60'for each variety and process (step S21-1, FIG. 4). In the seedling discrimination system 2, unlike the seedling data generation system 1, the seedling to be imaged does not necessarily have to be one seedling determined to be “good to grow”. Here, as described above, since the flow in which the seedling storage data generated by the seedling data generation system 1 is not used as the teacher data is described, the seedlings are not limited to the seedlings determined to be “good to grow”.
Then, the cultivar name and the process name of the selected seedling 60'are input by the operator via the input means 31 (step S21-2). For example, the cultivar name "Reiyo" and the process name "initial seedling raising" of "tomato" are input.

Next, the whole image of the selected seedling 60'is imaged by the image pickup means 40 as an image having colors (steps S21-3, FIG. 4, FIG. 5 (A)). Reference numeral 70'in FIG. 5A indicates image data captured by the imaging means 40. Needless to say, the image pickup means 40 should be able to capture and acquire the entire image of the seedling 60'to be imaged, that is, information representing all directions of the seedling appearance (for example, a three-dimensional image).
In the embodiment, the cultivar name and the process name are input in step S21-2, and then the seedlings are imaged in step S21-3, but the order of steps S21-2 and 3 may be reversed. ..

The variety name (“Reiyo” of “Tomato”) and the process name (“Initial seedling raising”) input in step S21-2 by the seedling basic data acquisition means 110, and the image data 70 of the seedlings imaged by the image pickup means 40. 'Is linked to the seedling basic data (step S21-4).
In this way, the seedling basic data is acquired through steps S21-1 to S21-4.

Next, when the image data 70'of the seedling basic data is image-processed, it is performed in the following steps (steps S21-5 to 7).
First, the three-dimensional image 70'of the seedling imaged by the image pickup means 40 is processed so that data is sent from the image pickup means to the control means 11 and the color is removed by the image processing means 111 to be expressed only in black and white. (Step S21-5, FIG. 2, FIG. 5 (B)). Reference numeral 71'in FIG. 5B indicates a three-dimensional image in which the captured seedling 60'is represented in black and white.

Then, the image data 71'of the seedlings represented in black and white is further processed by the image processing means 111 so as to be represented only by the contour line (step S21-6, FIG. 5C). Reference numeral 72'in FIG. 5C indicates image data represented only by contour lines.
Further, the image data 72'represented only by the contour line is overwritten with the image data of the seedling basic data. Then, this seedling basic data is stored in the storage means 12 as the teacher data 124 of the seedling discrimination system (step S21-7), and step S21 ends.

The machine learning means 115 acquires teacher data (seedling basic data) 124, and performs machine learning (supervised learning) to determine whether or not seedlings are well grown. Then, the teacher data (seedling basic data) including the variety name, the process name, and the image data is associated with the determination result of whether or not the seedlings are well grown, and this is also stored in the storage means 12 as the teacher data 124. , Create a database (step S22).

As the "index for determining that the growth is good", the area (size) of the first true leaf can be mentioned as in the seedling data generation system 1. The machine learning means 115 recognizes the seedling part (first true leaf) and the numerical range (area) of the part by machine learning, and automatically determines whether or not the growth is good. The position of the first true leaf is recognized, for example, as a range of coordinate values in the space of the image data 72 of the seedling.

Another "index for determining good growth" is, for example, the presence or absence of a seedling growth point (not shown). The growth point is the part where the plant newly grows buds due to cell division. For example, in FIG. 4, the part 60'-1 in the middle of the stem where the third true leaf extends, that is, between the extension of the stem and the shoot apex that produces new leaves, corresponds to the growth point of the seedling.
The growth point has a color different from that of the surrounding stems and the like, and the growth point can be specified by detecting this color. Seedlings without a growth point cannot continue to grow and are considered to be poorly grown seedlings. Since such poorly grown seedlings do not have growth points in the first place, no growth points are detected even by image processing. In addition, where one bud (stem) needs to grow from the growth point, it is considered to be a malformed seedling in which a plurality of buds grow. Such malformed seedlings are detected with a plurality of growth points.
Therefore, when the image data 70'of the captured seedling is sent from the image pickup means 40 to the control means 11, the image processing means 111 of the control means detects a point of a specific color from the image data of the seedling. When the color point is not detected, the image processing means 111 transmits to that effect to the control means 11, and the seedling is determined to be "not well grown (of the seedling)". When the growing point is detected, the image processing means 111 counts the number. If the number of seedlings is one, it is determined that "(seedling) growth is good", and if it is two or more, it is determined that "(seedling) growth is not good".

4.2 Processing procedure of seedling discrimination system (when seedling memory data is used as teacher data)
Next, in the seedling discrimination system 2, when the seedling storage data generated by the seedling data generation system 1 is used as teacher data when "growth is good", that is, the seedling data generation system and the seedling discrimination system are made continuous. The flow of the case is described.
When the seedling storage data generated by the seedling data generation system 1 described in 4.1 above is not used as teacher data when "growth is good", the seedling discrimination system 2 itself acquires seedling basic data or an image. Black-and-white and contour image processing of the data was performed (FIGS. 7A and 7B), and the data including the image data showing only the contour was used as the seedling basic data and used as the teacher data for machine learning. However, when the seedling storage data of the seedling data generation system 1 is used as teacher data when "growth is good", the seedling data generation system 1 acquires basic seedling data, sets parameters, and generates seedling storage data. The seedling discrimination system 2 differs in that the seedling storage data generated by the seedling data generation system 1 is used as teacher data for machine learning.

FIG. 8 shows a flow chart for explaining the flow of generating seedling memory data and determining the quality of seedling growth. When the seedling data generation system 1 and the seedling discrimination system 2 are continuous, that is, the seedling data generation system 1 performs the seedling basic data acquisition step, the parameter setting step, and the seedling storage data generation step. No. 2 includes a seedling storage data acquisition step and a machine learning step in order until the seedling growth is determined.
Further, even when the seedling memory data is used as the teacher data, the same reference code is attached to the common components and the description thereof is omitted, and the seedling memory data is used as in the case where the seedling memory data is not used as the teacher data. The configuration different from the case where the teacher data is not used will be mainly described. FIGS. 1, 2, 4 to 6 are adopted as they are even when the seedling storage data is used as the teacher data.

Even when the seedling storage data generated by the seedling data generation system 1 is used as teacher data when "growth is good", the flow of seedling storage data generation in the seedling data generation system 1 (steps S11, S12 (steps S11, S12). S12-1 to 7) and S13) are the same (FIGS. 3 (A) and 3 (B), 8).
Roughly, only one seedling (seedling to be raised) 60 determined to be “good to grow” for each variety and process was selected by the worker (step S11-1, FIG. 4), and the variety name and the process name were changed to It is input by the operator via the input means 31 (step S11-2). The seedlings to be raised are imaged in three dimensions with colors (step S11-3, FIG. 4), and this image data 70 is associated with the variety name and the process name to be the seedling basic data (step S11-4). , FIG. 5 (A). The image data 70 of the seedling is processed so that the color is removed by the image processing means 111 and only black and white (step S11-5, FIG. 5B) is represented, and further, only the contour line is expressed (step S11-6). , FIG. 5 (C). Then, the image data 72 represented only by this contour line is overwritten with the image data of the seedling basic data (step S11-7).
For the image data 72 represented only by the contour line, the part of the seedling which is an index for determining that the growth is good and the numerical range of the part are set by the operator and stored in the recording means 12 as the parameter data 122. (Step S12). Then, the image processing means 111 changes the site designated as a parameter (for example, the first true leaf 61) within the designated numerical range, and new seedlings containing the new sites 61-1 to 61-6. Image data 72-1 to 72-6 are generated, and new image data of seedlings is stored in the storage means 12 as seedling storage data 123 (step S13, FIG. 5).

Subsequently, the machine learning means 115 in the seedling discrimination system 2 acquires the seedling storage data 123 stored in step S13 as teacher data, and performs machine learning (supervised learning) to determine whether or not the seedlings are well grown. .. Then, the teacher data (seedling storage data) including the variety name, the process name, and the image data is associated with the determination result of whether or not the seedlings are well grown, and this is also stored in the storage means 12 as the teacher data 124. Create a database (step S22).

4.3 Effect of seedling discrimination system According to the seedling discrimination system 2, it is possible to accurately grasp the whole image (seedling appearance) of the seedlings to be raised, as in the seedling data generation system 1. In addition, the seedling discrimination system 2 According to this, it is possible to automatically discriminate between "well-grown" seedlings and "poor-grown seedlings (immature seedlings and malformed seedlings)" by machine learning. Therefore, it is possible to appropriately determine the quality of the growth of the original seedlings and the grafted seedlings without being based on the experience of a skilled person.
In particular, as described in 4.2 above, if the seedling discrimination system 2 uses the seedling storage data 124 generated by the seedling data generation system 1 as teacher data, the image data of the seedlings when "growth is good" can be obtained. It is automatically collected, saving you the time of collecting teacher data.

Further, also in the seedling discrimination system 2, by performing image processing using only black and white and contour, it is possible to secure the data capacity of the storage means while suppressing only the data necessary for the quality judgment of breeding. Not to mention.

5. In addition, the above-mentioned examples are for explaining the present invention, and do not limit the present invention in any way, and all the modifications and modifications made within the technical scope of the present invention are included in the present invention. Needless to say, it will be done.

The seedling data generation system, seedling discrimination system, seedling data generation program, seedling discrimination system, seedling data generation device, and seedling discrimination device of the present invention exemplify, but are not limited to, tomato (vegetable) seedlings as examples. , Fruits, flowers, etc. can be widely applied.
Further, as an example, the original seedling in the process of initial seedling raising is exemplified, but the present invention is not limited to this. For example, it can be applied not only to initial seedling raising but also to original seedlings or grafted seedlings in processes such as germination management, plug shipping, potting, secondary seedling raising, sorting, and surplus management.

10 Seedling data generation device, seedling discrimination device 11 Control means 110 Seedling basic data acquisition means 111 Image processing means 112 Parameter setting means 113 Seedling storage data generation means 114 Teacher data generation means 115 Machine learning means 12 Storage means 31 Input means

Claims (7)

Seedling basics to acquire basic seedling data including image data showing the overall image of the seedlings to be raised, the varieties of the seedlings, and the name of one of the multiple steps until the raising of the seedlings is completed. Data acquisition means and
Seedling storage that generates seedling storage data by associating a parameter including at least one part of the seedling and a numerical range of the part with the image data of the seedling basic data acquired by the seedling basic data acquisition means. Equipped with data generation means
When the seedling site and the numerical range of the site, which are indicators for determining that the growth is good, are set as parameters for the seedling basic data of one seedling that is determined to be good for each variety and process. The seedling storage data generation means generates a plurality of image data of seedlings according to the parameters from the image data acquired by the seedling storage data generation means, adds them to the image data of the seedling basic data, and adds the seedling basic data. Seedling data generation system that generates seedling storage data from data. The seedling basic data acquisition means further includes an image processing means.
The seedling data generation system according to claim 1, wherein the image processing means removes colors from image data representing an overall image of seedlings and extracts only the outline of the seedlings. An imaging means that captures image data that represents the overall picture of the seedlings to be raised,
An input means for inputting the cultivar name of the seedling and the name of one of the plurality of steps until the seedling growth is completed.
A control means for determining whether or not the seedlings to be raised are good or not,
Equipped with a storage means to store data
The imaging means captures image data including three-dimensional data of one seedling determined to be good for growing for each seedling variety and process.
Further, in the input means, the part of the seedling, which is an index for determining that the growth is good, and the numerical range of the part are input as parameters.
The control means is an image processing means that generates a plurality of image data of seedlings according to the parameters from the image data captured by the image pickup means, and the input means for the image data generated by the image processing means. The seedling basic data is created by associating the variety name entered in step 1 with the name of one process and the determination result that the breeding is good, and the seedling basic data is added to the storage means as teacher data and stored. Have means and
Further, the control means performs learning using the seedling basic data as teacher data for discrimination, and provides a machine learning means for discriminating whether or not seedling growth is good from the image data generated by the image processing means. seedlings determination system having. Computer,
Seedling basics to acquire basic seedling data including image data showing the overall image of the seedlings to be raised, the varieties of the seedlings, and the name of one of the multiple steps until the raising of the seedlings is completed. Data acquisition means and
Seedling storage that generates seedling storage data by associating a parameter including at least one part of the seedling and a numerical range of the part with the image data of the seedling basic data acquired by the seedling basic data acquisition means. Data generation means and
It is a seedling data generation program to make it work.
When the seedling site and the numerical range of the site, which are indicators for determining that the growth is good, are set as parameters for the seedling basic data of one seedling that is determined to be good for each variety and process. The seedling storage data generation means generates a plurality of image data of seedlings according to the parameters from the image data acquired by the seedling storage data generation means, adds them to the image data of the seedling basic data, and adds the seedling basic data. Seedling data generation program that generates seedling storage data from data. Computer,
An imaging means that captures image data that represents the overall picture of the seedlings to be raised,
An input means for inputting the cultivar name of the seedling and the name of one of the plurality of steps until the seedling growth is completed.
A control means for determining whether or not the seedlings to be raised are good or not,
Equipped with a storage means to store data
The imaging means captures image data including three-dimensional data of one seedling determined to be good for growing for each seedling variety and process.
Further, in the input means, the part of the seedling, which is an index for determining that the growth is good, and the numerical range of the part are input as parameters.
The control means is an image processing means that generates a plurality of image data of seedlings according to the parameters from the image data captured by the image pickup means, and the input means for the image data generated by the image processing means. The seedling basic data is created by associating the variety name entered in step 1 with the name of one process and the determination result that the breeding is good, and the seedling basic data is added to the storage means as teacher data and stored. Have means and
Further, the control means performs learning using the seedling basic data as teacher data for discrimination, and provides a machine learning means for discriminating whether or not seedling growth is good from the image data generated by the image processing means. seedlings discrimination program having. Seedling basics to acquire basic seedling data including image data showing the overall image of the seedlings to be raised, the varieties of the seedlings, and the name of one of the multiple steps until the raising of the seedlings is completed. Data acquisition means and
Seedling storage that generates seedling storage data by associating a parameter including at least one part of the seedling and a numerical range of the part with the image data of the seedling basic data acquired by the seedling basic data acquisition means. Equipped with data generation means
When the seedling site and the numerical range of the site, which are indicators for determining that the growth is good, are set as parameters for the seedling basic data of one seedling that is determined to be good for each variety and process. The seedling storage data generation means generates a plurality of image data of seedlings according to the parameters from the image data acquired by the seedling storage data generation means, adds them to the image data of the seedling basic data, and adds the seedling basic data. A seedling data generator that generates seedling storage data from data. An imaging means that captures image data that represents the overall picture of the seedlings to be raised,
An input means for inputting the cultivar name of the seedling and the name of one of the plurality of steps until the seedling growth is completed.
A control means for determining whether or not the seedlings to be raised are good or not,
Equipped with a storage means to store data
The imaging means captures image data including three-dimensional data of one seedling determined to be good for growing for each seedling variety and process.
Further, in the input means, the part of the seedling, which is an index for determining that the growth is good, and the numerical range of the part are input as parameters.
The control means is an image processing means that generates a plurality of image data of seedlings according to the parameters from the image data captured by the image pickup means, and the input means for the image data generated by the image processing means. The seedling basic data is created by associating the variety name entered in step 1 with the name of one process and the determination result that the breeding is good, and the seedling basic data is added to the storage means as teacher data and stored. Have means and
Further, the control means performs learning using the seedling basic data as teacher data for discrimination, and provides a machine learning means for discriminating whether or not seedling growth is good from the image data generated by the image processing means. seedling discriminating apparatus having.

Images