JP7677633B2 - State estimation system, state estimation method, feed amount determination system, feed amount determination method, trained model generation device, and trained model generation method - Google Patents

Description

The present invention relates to a state estimation system, a state estimation method, a feeding amount determination system, a feeding amount determination method, a trained model generation device, and a trained model generation method.

Due to the aging of dairy farmers and a lack of successors, the number of dairy farms and the number of cows kept in Japan is on the decline. On the other hand, the number of cows kept per farm is increasing, dairy farms are becoming larger in scale, and the amount of milk produced per cow is also increasing. In order to maintain a stable amount of raw milk production in the future, it is hoped that new farmers will enter the industry.

Feed design is one of the important aspects of dairy cow feeding management, and the Japanese Feeding Standards and the NRC (National Research Council) Feeding Standards stipulate the amount of feed to be given. However, in pasture feeding management, cows are allowed to freely graze on pasture grass, making it difficult to grasp the amount of grass eaten and designing an appropriate amount of supplementary feed to be given. Furthermore, from an economic perspective, it is desirable to avoid giving excessive feed, but if the amount of feed given is not appropriate, there is a risk of the cows becoming malnourished and losing milk productivity.

Experienced dairy farmers may be able to judge the condition of cows by their abdominal turgor and the color and luster of their bodies, but this can be difficult for new farmers with little knowledge or experience. For this reason, there is a demand for the development of technology to grasp the nutritional status of cows and manage their upbringing without relying on the experience of dairy farmers.

Patent documents 1 to 4 describe technologies for understanding the feeding behavior and feeding amount of cattle based on the amount of walking or jaw movement of the cattle, and for use in feeding management. In addition, patent document 5 describes a technology for monitoring cattle behavior in real time, which detects cattle births, movement outside of a management area, etc., from cattle movement trajectory data.

Japanese Patent Application Publication No. 10-213655 Japanese Patent Application Publication No. 10-262498 JP 2002-262712 A JP 2007-173930 A Japanese Patent Application Publication No. 10-160820

The nutritional status of cattle is influenced not only by the feeding behavior and amount of feed intake of cattle, but also by the rearing environment, such as the size of the pasture, the type of grass, and the component content of the grass. With rearing management based on the feeding behavior and amount of feed intake of cattle, as in the techniques described in Patent Documents 1 to 4, it is difficult to accurately grasp the nutritional status of cattle, and appropriate rearing management may not be possible. In addition, the technology described in Patent Document 5 is a technology that detects the birth of cattle or movement outside the management area by monitoring the behavior of cattle in real time, but is not a technology that estimates the nutritional status of cattle.

In other words, one aspect of the present invention was made in consideration of the above problems, and its purpose is to provide a technology for estimating the condition of livestock, such as their nutritional state or health state.

To solve the above problem, a state estimation system according to one aspect of the present invention includes an estimation unit that uses position trajectory data representing the position trajectory of livestock as input to estimate the state of the livestock, using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of the livestock and state data representing the state of the livestock as training data.

A state estimation method according to one aspect of the present invention includes an estimation step of estimating the state of livestock by using position trajectory data representing the position trajectory of livestock as input, using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of livestock and state data representing the state of the livestock as training data.

The feed amount determination system according to one aspect of the present invention includes a feed amount determination unit that determines the amount of feed to be given to the livestock by referring to the state of the livestock estimated by the state estimation system.

The feed amount determination method according to one aspect of the present invention includes a feed amount determination step for determining the amount of feed to be fed to the livestock by referring to the state of the livestock estimated by the above-mentioned state estimation method.

A trained model generation device according to one aspect of the present invention includes a model generation unit that performs machine learning using position trajectory data representing the position trajectory of livestock and status data representing the status of the livestock as training data, and generates a trained model that estimates the status of the livestock using position trajectory data representing the position trajectory of the livestock as input.

A trained model generation method according to one aspect of the present invention includes a model generation step of performing machine learning using position trajectory data representing the position trajectory of livestock and status data representing the status of the livestock as training data to generate a trained model that estimates the status of the livestock using the position trajectory data representing the position trajectory of the livestock as input.

The state estimation system, feeding amount determination system, and trained model generation device according to each aspect of the present invention may be realized by a computer. In this case, the control programs for the state estimation system, feeding amount determination system, and trained model generation device, which realize the state estimation system, feeding amount determination system, and trained model generation device on a computer by causing a computer to operate as each part (software element) of these systems and devices, as well as computer-readable storage media on which they are stored, also fall within the scope of the present invention.

According to one aspect of the present invention, it is possible to estimate the condition of livestock, such as their nutritional state or health state.

1 is a block diagram showing a configuration of a main part of a nutritional state estimation system according to one embodiment of the present invention. FIG. 2 is a diagram illustrating an example of position trajectory data generated in the nutritional state estimation system according to one aspect of the present invention. FIG. 1 is a diagram illustrating an example of teacher data used in a trained model generation device according to one embodiment of the present invention. FIG. 2 is a flow chart showing the flow of a nutritional state estimation process in the nutritional state estimation system according to one embodiment of the present invention. FIG. 1 is a flow chart showing the flow of a learning process in a trained model generation device according to one embodiment of the present invention. FIG. 1 is a block diagram showing the configuration of a main part of a feed amount determination system according to one embodiment of the present invention.

The inventors discovered that there is a correlation between the position trajectories of livestock and the condition of the livestock, and that by using a trained model that has learned this correlation through machine learning, it is possible to estimate the condition of the livestock from position trajectory data that represents the position trajectories of the livestock, leading to the completion of the present invention.

Examples of the state of livestock that can be estimated by a state estimation system according to one embodiment of the present invention include, but are not limited to, nutritional state, health state, and estrus state, as well as combinations of these. When the state estimation system according to one embodiment of the present invention estimates the nutritional state of livestock as the state of livestock, the state estimation system can be called a nutritional state estimation system. As an example, an embodiment of the present invention in which the state estimation system is a nutritional state estimation system will be described below.

The nutritional state estimation system according to one aspect of the present invention includes an estimation unit that uses position trajectory data representing the position trajectory of livestock as input to estimate the nutritional state of the livestock, using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of the livestock and nutritional state data representing the nutritional state of the livestock as training data.

The subjects for which the nutritional state is estimated in the nutritional state estimation system are livestock such as cows, pigs, horses, goats, and sheep. Cattle, including dairy cows and beef cattle, are particularly suitable for estimating the nutritional state of animals using the nutritional state estimation system, as they are raised on a large scale for industrial purposes. There are no particular limitations on the type of livestock rearing, and they may be reared either indoors or outdoors.

[Nutritional State Estimation System 100]
1 is a block diagram showing a configuration of a main part of a nutritional state estimation system 100 according to one aspect of the present invention. The nutritional state estimation system 100 includes a nutritional state estimation device 20. The nutritional state estimation system 100 further includes a user terminal 10, a data generation device 30, and a trained model generation device 40.

The user terminal 10, the nutritional state estimation device 20, and the data generation device 30 may be implemented in a manner communicatively connected to each other via a communication means, or may be implemented integrally. The communication means may be a LAN (Local Area Network) such as a wired LAN or a wireless LAN, or may be a WAN (Wide Area Network).

(User terminal 10)
The user terminal 10 mediates an input signal from a user to at least one of the nutritional state estimation device 20, the data generation device 30, and the trained model generation device 40. In addition, the user terminal 10 mediates an output signal from at least one of the nutritional state estimation device 20, the data generation device 30, and the trained model generation device 40 to the user.

The user terminal 10 includes at least one of an operation unit (not shown) that mediates input signals and a display unit (not shown) that mediates output signals. The operation unit may be implemented in the form of an operation device such as a keyboard, a mouse, or a touch panel. The display unit may be implemented in the form of a display device such as a display, or a printing device such as a printer. The operation unit and the display unit may be implemented integrally in the form of an information processing device such as a smartphone or a personal computer.

(Nutritional state estimation device 20)
The nutritional state estimation device 20 estimates the nutritional state of a livestock from position trajectory data representing the position trajectory of the livestock. The nutritional state estimation device 20 includes a control unit 21 and a storage unit 22. The nutritional state estimation device 20 may be implemented in the form of an information processing device such as a personal computer.

The control unit 21 controls each part of the nutritional state estimation device 20, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 21. As each part, the control unit 21 includes an acquisition unit 23 and an estimation unit 24.

The acquisition unit 23 acquires position trajectory data representing the position trajectory of livestock. The acquisition unit 23 may be implemented, for example, in the form of a wireless or wired communication interface connected to the control unit 21. The acquisition unit 23 acquires input data from at least one of the user terminal 10, the memory unit 22, and the data generation device 30. The acquisition unit 23 transmits the acquired input data to the estimation unit 24. The processing performed by the acquisition unit 23 is controlled by a processor provided in the control unit 21.

The estimation unit 24 estimates the nutritional state of the livestock by using the trained model generated by performing machine learning using the position trajectory data representing the position trajectory of the livestock and the nutritional state data representing the nutritional state of the livestock as training data. The estimation unit 24 may be implemented, for example, on a memory provided in the control unit 21. As an example, the estimation unit 24 estimates the nutritional state of the livestock by using the trained model generated by the trained model generation device 40 and the position trajectory data of the livestock as input. The estimation unit 24 transmits the estimated nutritional state data representing the nutritional state of the livestock to at least one of the user terminal 10 and the memory unit 22. The processing performed by the estimation unit 24 is controlled by a processor provided in the control unit 21.

The storage unit 22 stores data necessary for processing in the nutritional state estimation device 20. The storage unit 22 is, for example, a hard disk drive (HDD) and a read only memory (ROM).
The control unit 21 may be implemented in the form of a non-volatile memory such as a statically linked memory (SRAM) or a statically linked only memory (SRAM). The processing executed by the storage unit 22 is controlled by a processor included in the control unit 21. The storage unit 22 stores data of a program for implementing the processing in the nutritional state estimation device 20. In addition, the storage unit 22 stores data received from other components in the nutritional state estimation system 100. The storage unit 22 stores the trained model generated by the trained model generation device 40.

(Data Generating Device 30)
The data generating device 30 generates position trajectory data. The data generating device 30 includes a control unit 31. The data generating device 30 may be implemented in the form of an information processing device such as a personal computer.

The control unit 31 controls each part of the data generating device 30, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 31. As each part, the control unit 31 includes an acquisition unit 32 and a data generating unit 33.

The acquisition unit 32 acquires location information of the livestock. The acquisition unit 32 may be implemented, for example, in the form of a wireless or wired communication interface connected to the control unit 31. As an example, the acquisition unit 32 acquires location information of the livestock measured by a measuring device (not shown). The processing performed by the acquisition unit 32 is controlled by a processor provided in the control unit 31.

Examples of measuring devices include: combinations of transmitters and receivers of positioning systems such as GNSS (Global Navigation Satellite System) and WLAN (Wireless Local Area Network); imaging devices such as cameras and video cameras; acceleration sensors; and the like. In combinations of transmitters and receivers of positioning systems, the transmitters are typically attached to livestock, and the receivers are communicably connected to an information processing device. In the form of imaging devices, the imaging devices are typically installed so as to be able to image the livestock. In the form of acceleration sensors, the acceleration sensors are typically attached to the livestock. The measuring device can be appropriately selected depending on the type of livestock and the way it is kept. It may be preferable for the measuring device to be an imaging device, since it is capable of non-invasive measurement.

The data generation unit 33 generates position trajectory data by referring to the livestock position information acquired by the acquisition unit 32 over a predetermined period of time. The data generation unit 33 may be implemented, for example, in the form of a combination of a storage medium such as a memory and a processor that executes a program stored in the storage medium. The data generation unit 33 transmits the generated position trajectory data to the nutritional state estimation device 20 or the trained model generation device 40.

The data generating unit 33 generates position trajectory data by referring to changes over time in the position information of the livestock within a predetermined period of time using a conventionally known method. As an example, the data generating unit 33 may generate image data in which the position trajectory of the livestock is projected onto a two-dimensional plane, i.e., position trajectory data, by applying at least one of an object detection algorithm and a skeletal detection algorithm capable of detecting characteristic points of the livestock's body to a moving image of the livestock captured using an imaging device (not shown). As an example of an object detection algorithm, Yolo (You
Examples of the skeleton detection algorithm include DeepLabCut, OpenPose, DeepPose, and DeeperCUT.

<Position trajectory data>
The position trajectory data is data representing the position trajectory of the livestock, including the position information of the livestock continuously acquired for a predetermined period. The position trajectory data is generated for each individual livestock. As used herein, the term "position trajectory of the livestock" means a path through which the feature points of the livestock's body have passed within a predetermined period. Examples of the feature points of the livestock's body include parts of the livestock's body, such as bones, joints, parts, and tips of parts; the center of gravity of the livestock's body; signs attached to the livestock's body; combinations of these; and the like. Examples of the signs attached to the livestock's body include a transmitter that can be used in a positioning system and an acceleration sensor. The feature points of the livestock's body may be one or a combination of multiple feature points. The user can appropriately select these feature points according to the desired information. In addition, the predetermined period for which the position information is acquired is not particularly limited, but may be 3 hours, 6 hours, 10 hours, 12 hours, 24 hours, 36 hours, 48 hours, and the like.

There are no particular limitations on the position trajectory data, so long as it allows a person skilled in the art to uniquely restore the position trajectory of livestock by referring to the position trajectory data using a conventionally known method. Examples of position trajectory data include: image data representing the position trajectory of livestock; numerical data representing the correspondence between the position of a feature point and time; text data explaining the direction and amount of movement of a feature point; and combinations of these. Among these data, image data representing the position trajectory of livestock is preferred because the data is easy to acquire, process, and manage.

Here, an example of position trajectory data generated by the data generation unit 33 will be described with reference to FIG. 2. FIG. 2 is a diagram illustrating an example of position trajectory data generated in a nutritional state estimation system according to one embodiment of the present invention. In FIG. 2, RD1 is a schematic diagram showing a livestock cow grazing freely in a pasture, overlaid with a black line indicating the path traveled by the characteristic points of the cow's body during a specified period of time. As shown in FIG. 2, the data generation unit 33 generates position trajectory data TD1, which is image data representing the path traveled by the characteristic points of the cow's body during a specified period of time.

Image data representing the position trajectories of livestock can be generated using conventionally known image processing technology. It may be two-dimensional image data in which the position trajectories of livestock are projected onto a two-dimensional plane, or it may be three-dimensional image data in which the position trajectories of livestock are transferred onto a three-dimensional space. The image data may be monochrome image data or color image data. The position trajectory data may be data that has been subjected to corrections such as conversion and compression, as long as the effect of the present invention is not impaired.

(Trained model generation device 40)
The trained model generation device 40 receives position trajectory data representing the position trajectory of livestock and generates a trained model for estimating the nutritional state of the livestock. The trained model generation device 40 includes a control unit 41. The trained model generation device 40 may be implemented in the form of an information processing device such as a personal computer.

The control unit 41 controls each part of the trained model generation device 40, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 41. As each part, the control unit 41 includes an acquisition unit 42 and a model generation unit 43.

The acquisition unit 42 acquires teacher data. The acquisition unit 42 acquires the position trajectory data generated by the data generation device 30, and generates teacher data by associating the position trajectory data with nutritional state data that indicates the nutritional state of the livestock that is associated with the position trajectory data. The acquisition unit 42 transmits the generated teacher data to the model generation unit 43.

<Nutritional Status Data>
Nutritional status data is data that represents the nutritional status of livestock. As used herein, the term "nutritional status of livestock" refers to information regarding at least one of the types and amounts of nutrients ingested by livestock. Types of nutrients include, for example, substances such as proteins, carbohydrates, lipids, fibers, vitamins, and minerals; parameters that can be calculated from these substances; and the like. Parameters include, for example, calories, total digestible nutrients (TDN), and the like.

The nutritional state data may be data representing a nutritional sufficiency rate of the livestock. As an example, the nutritional sufficiency rate of the livestock may be a TDN sufficiency rate calculated by the following formula (1):
TDN sufficiency rate = TDN supply / TDN requirement (1)
In formula (1), the TDN supply amount is the product of the amount of feed consumed by the livestock and the TDN contained in the unit amount of the feed. The TDN requirement amount is, for example, the TDN required for the livestock to carry out normal production activities, which is set according to the weight and birth frequency of the livestock.

Examples of nutritional status data include numerical data (sufficiency rate, nutrient amount, etc.) that indicate the nutritional status of livestock, and nutritional status class data (high, low, etc.). Nutritional status data can be generated for each individual livestock.

<Teacher data>
Here, the teacher data generated by the acquisition unit 42 will be described with reference to FIG. 3. FIG. 3 is a diagram for explaining an example of teacher data used in a trained model generation device according to one aspect of the present invention. In FIG. 3, the teacher data LD1 is teacher data including number data, position trajectory data, and nutritional state data. The number data is label data that associates the position trajectory data with the nutritional state data. The number of the number data typically corresponds to one individual of livestock. In FIG. 3, the position trajectory data is image data representing the position trajectory of the livestock. The image data is composed of black lines and white margins. The black lines represent the position trajectory of the livestock projected onto a two-dimensional plane. The white margin represents an area projected onto a two-dimensional plane through which the feature points of the livestock's body did not pass. In FIG. 3, the nutritional state data is class data determined according to the TDN sufficiency rate. As an example, the nutritional state of livestock with a TDN sufficiency rate of 100% or more is classified into a "high" class, and the nutritional state of livestock with a TDN sufficiency rate of less than 100% is classified into a "low" class.

The trained model used in this embodiment is not limited to the trained model constructed using the training data LD1 exemplified in FIG. 3. The training data may include, for example: data on livestock such as weight, sex, age, and birth frequency; data on the rearing environment such as pasture area, type of pasture, weather, and rearing style; etc.

The model generation unit 43 uses the teacher data acquired by the acquisition unit 42 to generate a trained model that receives position trajectory data representing the position trajectory of livestock and outputs the nutritional state of the livestock. The trained model is a trained model that has been machine-learned to determine the correlation between the position trajectory data and the nutritional state data. The model generation unit 43 may be implemented, for example, on a memory provided in the control unit 41. The processing performed by the model generation unit 43 is controlled by a processor provided in the control unit 41. The model generation unit 43 transmits the generated trained model to the nutritional state estimation device 20.

The trained model can be constructed by a conventionally known machine learning method. Examples of machine learning methods include neural networks such as convolutional neural networks and recurrent neural networks; trees such as decision trees and random forests; regression such as support vector machines; stochastic gradient descent; and the like. The trained model may be constructed by a transfer learning method that applies an existing trained model. As an example, the trained model is constructed by applying a trained model constructed by learning data of a "high" class of livestock nutritional status to learning data of a "low" class of livestock nutritional status. In other words, the trained model may be constructed by applying a model constructed using data of a class with a large number of samples to learning a class with a small number of samples, using a transfer learning method.

(Nutrition status estimation process flow)
The flow of the nutritional state estimation process (nutritional state estimation method) executed by the nutritional state estimation device 20 of the nutritional state estimation system 100 will be described with reference to Fig. 4. Fig. 4 is a flow diagram showing the flow of the nutritional state estimation process in the nutritional state estimation system 100 according to one aspect of the present invention.

First, in step S11, the acquisition unit 23 acquires the position trajectory data generated by the data generation device 30. The acquisition unit 23 transmits the acquired position trajectory data to the estimation unit 24. Next, the estimation unit 24 acquires the trained model generated by the trained model generation device 40. Then, the estimation unit 24 inputs the position trajectory data into the trained model and acquires nutritional state data (step S12, estimation step).

Here, the input data to be input to the trained model may include other data besides the position trajectory data, as long as the effect of the present invention is not impaired. Examples of other data include data on livestock, such as weight, sex, age, and number of births; data on the rearing environment, such as pasture area, type of pasture, weather, and rearing style; etc.

The estimation unit 24 outputs the output nutritional state data as an estimation result to the user terminal 10 (step S13), and the nutritional state estimation process ends. Note that, when the nutritional state data is numerical data such as a sufficiency rate, the estimation unit 24 may determine whether the nutritional state of the livestock is high or low from the numerical data, and output the determination result to the user terminal 10.

(Flow of trained model generation process)
The flow of a trained model generation process (trained model generation method) executed by the trained model generation device 40 will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing the flow of a learning process in the trained model generation device according to one aspect of the present invention.

First, in step S21, the acquisition unit 42 acquires teacher data LD1 including position trajectory data and corresponding nutritional state data. The acquisition unit 42 transmits the acquired teacher data LD1 to the model generation unit 43. Next, the model generation unit 43 uses the acquired teacher data LD1 to generate a trained model that takes the position trajectory data as input and outputs nutritional state data (step S22, model generation step).

The model generation unit 43 transmits the generated trained model to the nutritional status estimation device 20 (step S22), and the trained model generation process ends.

As described above, according to the nutritional state estimation system of one aspect of the present invention, the nutritional state of livestock can be estimated using a trained model that estimates the nutritional state of livestock by inputting position trajectory data representing the position trajectory of the livestock. According to the nutritional state estimation system of one aspect of the present invention, the nutritional state of livestock can be accurately estimated simply by acquiring the movement trajectory of the livestock, without taking into account the feeding behavior and feeding amount of the livestock, the rearing environment of the livestock, etc. According to the nutritional state estimation system of one aspect of the present invention, since the knowledge or experience of the user is not required to estimate the nutritional state of livestock, it can also be suitably used for breeding management by new farmers with little knowledge or experience.

In addition, a trained model generated for one farm or livestock group can be used to estimate the nutritional status of other farms or livestock groups, and a generalized trained model can be provided. Furthermore, image data that represents the position trajectory of livestock within a specified period as a single image can be used as the position trajectory data used by the nutritional status estimation system, making it easy to acquire, process, and manage the data.

[Feeding amount determination system]
The feed amount determination system according to one embodiment of the present invention includes a feed amount determination unit that determines the amount of feed to be fed to livestock by referring to the nutritional state of the livestock estimated by the nutritional state estimation system 100 according to one embodiment of the present invention described above.

Figure 6 is a block diagram showing the main components of a feeding amount determination system 200 according to one embodiment of the present invention. As shown in Figure 6, the feeding amount determination system 200 includes a feeding amount determination device 50 along with the nutritional state estimation system 100 of Figure 1. The user terminal 10, nutritional state estimation device 20, data generation device 30, and trained model generation device 40, and the feeding amount determination device 50 may be implemented in a manner connected to each other so as to be able to communicate with each other via communication means, or may be implemented as an integrated unit.

(Feeding amount determination device 50)
The feed amount determination device 50 determines the amount of feed to be fed to the livestock by referring to the estimated nutritional state of the livestock. The feed amount determination device 50 includes a control unit 51. The feed amount determination device 50 may be implemented in the form of an information processing device such as a personal computer.

The control unit 51 controls each part of the feeding amount determination device 50, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 51. As each part, the control unit 51 has an acquisition unit 52 and a feeding amount determination unit 53.

The acquisition unit 52 acquires data on the nutritional state of the livestock estimated by the nutritional state estimation device 20. The acquisition unit 52 may be implemented, for example, in the form of a wireless or wired communication interface. The processing executed by the acquisition unit 52 is controlled by a processor provided in the control unit 51. The acquisition unit 52 transmits the acquired nutritional state data to the feed amount determination unit 53.

The feed amount determination unit 53 determines the amount of feed to be given to the livestock by referring to the estimated nutritional state of the livestock. The feed amount determination unit 53 may be implemented, for example, in a memory provided in the control unit 51. The processing performed by the feed amount determination unit 53 is controlled by a processor provided in the control unit 51.

The method by which the feed amount determination unit 53 determines the amount of feed to be given to the livestock by referring to the data on the estimated nutritional state of the livestock may be a conventionally known method, and the user can select it as appropriate. For example, one such method is a method of applying negative feedback using the estimated nutritional state and a predetermined criterion.

The feeding amount determination unit 53 transmits data on the determined feeding amount to at least one of a memory unit (not shown) provided in the feeding amount determination device 50 and the user terminal 10.

As described above, the feed amount determination system according to one embodiment of the present invention determines the amount of feed to be given to livestock by referring to the estimated nutritional state of the livestock. This allows the feed amount determination system according to one embodiment of the present invention to appropriately manage the livestock's livestock rearing, including determining the amount of feed to be given, according to the livestock's nutritional state. As a result, it is expected that the quality of products derived from livestock will improve and feeding costs will be reduced.

The feed amount determination method according to one aspect of the present invention includes a feed amount determination step of determining the amount of feed to be fed to the livestock by referring to the nutritional state of the livestock estimated by the nutritional state estimation method according to one aspect of the present invention. That is, an example of a system that executes the feed amount determination method according to one aspect of the present invention is the feed amount determination system 200. Therefore, details of the feed amount determination method according to one aspect of the present invention are as described above for the feed amount determination system according to one aspect of the present invention.

[Software implementation example]
The control blocks (particularly the control unit 21, control unit 31, control unit 41, and control unit 51) of the nutritional status estimation device 20, the data generation device 30, the learned model generation device 40, and the feeding amount determination device 50 may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or by software.

In the latter case, the nutritional state estimation device 20, the data generation device 30, the trained model generation device 40, and the feeding amount determination device 50 are equipped with a computer that executes instructions of a program, which is software that realizes each function. This computer is equipped with, for example, one or more processors, and a computer-readable recording medium that stores the above program. In the computer, the processor reads the above program from the recording medium and executes it, thereby achieving the object of the present invention. The above processor can be, for example, a CPU (Central Processing Unit). The above recording medium can be a "non-transient tangible medium", such as a ROM (Read Only Memory), tape, disk, card, semiconductor memory, programmable logic circuit, etc. In addition, a RAM (Random Access Memory) that expands the above program may also be included. The above program may be supplied to the computer via any transmission medium (such as a communication network or broadcast waves) that can transmit the program. Note that one aspect of the present invention can also be realized in the form of a data signal embedded in a carrier wave, in which the above program is embodied by electronic transmission.

[Modifications]
The nutritional state estimation device system may be realized as a nutritional state estimation device including the nutritional state estimation device and the trained model generation device as an integrated device. Furthermore, the nutritional state estimation device system may be realized as a nutritional state estimation device including the nutritional state estimation device, the data generation device, and the trained model generation device as an integrated device.

The present invention is not limited to the above-mentioned aspects. For example, one aspect of the present invention may be a health state estimation system that estimates the health state of livestock by using health state data representing the health state of livestock instead of using nutritional state data representing the nutritional state of livestock as teacher data in the above-mentioned nutritional state estimation system. Furthermore, for example, one aspect of the present invention may be an estrus state estimation system that estimates the estrus state of livestock by using estrus state data representing the estrus state of livestock instead of using nutritional state data representing the nutritional state of livestock as teacher data in the above-mentioned nutritional state estimation system. These aspects are due to the existence of a correlation between the state of livestock and the position trajectory of the livestock. That is, the state estimation system according to one aspect of the present invention includes an estimation unit that estimates the state of the livestock by using the position trajectory data representing the position trajectory of the livestock as an input, using a trained model generated by performing machine learning using the position trajectory data representing the position trajectory of the livestock and the state data representing the state of the livestock as teacher data. It is preferable that the state of the livestock estimated by the state estimation system is the nutritional state of the livestock or the health state of the livestock.

When the state estimation system according to one aspect of the present invention is a health state estimation system, the health state estimation system includes an estimation unit that estimates the nutritional state of the livestock by inputting position trajectory data representing the position trajectory of the livestock, using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of the livestock and health state data representing the health state of the livestock as training data. Examples of the health state data include data representing the medical history of the livestock, data representing the blood components of the livestock, and data representing general components contained in raw milk obtained from the livestock. Examples of the medical history include medical history for diseases such as mastitis, milk fever, loss of appetite, ketosis, and hoof disease.

When the state estimation system according to one aspect of the present invention is an estrus state estimation system, the estrus state estimation system includes an estimation unit that estimates the estrus state of the livestock by inputting position trajectory data representing the position trajectory of the livestock, using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of the livestock and estrus state data representing the estrus state of the livestock as training data. Examples of estrus state data include sex hormone concentrations in the blood and milk of the livestock, as well as the amount of activity and standing behavior of the livestock collected by various sensors such as a pedometer and an accelerometer.

The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in the different embodiments are also included in the technical scope of the present invention.

One embodiment of the present invention is described below.

Example 1
In this example, a trained model was generated by machine learning the correlation between the position trajectory data of a cow and the nutritional state data of the cow. The trained model was input with the position trajectory data, and output data was used to perform estimation, and the accuracy of the estimated data was evaluated.

[Generation of training data]
Four cows (dairy cows) were randomly assigned to the first group and four cows (dairy cows) to the second group. Each cow in the first group was given an amount of feed that gave a TDN sufficiency of 110%. Each cow in the second group was given an amount of feed that gave a TDN sufficiency of 80% to 90%. After feeding, a GNSS transmitter was attached to each cow in each group, and the cows were grazed in a pasture (dimensions: 90 m x 250 m) for one day. The position trajectory data of each cow was generated in the form of monochrome image data by referring to the data from the transmitter attached to each cow.
As nutritional state data, the position trajectory data of the first group of cattle was labeled as "high" and the position trajectory data of the second group of cattle was labeled as "low." In this way, training data was generated. The TDN sufficiency rate was calculated by the following formula (1):
TDN sufficiency rate = TDN supply / TDN requirement (1)
In formula (1), the TDN supply amount is the product of the amount of feed consumed by the livestock and the TDN contained in the unit amount of the feed, and the TDN requirement amount is the TDN required for the livestock to carry out normal metabolism, which is set according to the body weight and birth frequency of the livestock.

[Generating a trained model]
In the machine learning, an open software library "TensorFlow"("TensorFlow" is a registered trademark of Google, Inc.) was used. The trained model "label_image.py" of this embodiment was generated using "retrain.py", which is one of the transfer learning methods.

[Generating input data and obtaining estimation data]
Input data was generated by randomly selecting position trajectory data included in the training data. Input data was input to the trained model "label_image.py" to obtain numerical data of the likelihood classified into the "high" or "low" label as output data. The label with the higher likelihood of the "high" or "low" label was obtained as estimated data. The operations from generating input data to obtaining estimated data were performed five times independently.

[Dataset generation]
The operations from [generation of training data] to [generation of input data, acquisition of estimated data] were performed twice independently to generate two datasets (2019 and 2020). In addition, the data obtained by integrating the training data of each dataset was used as training data, and the operations from [generation of training data] to [generation of input data, acquisition of estimated data] were performed to generate one dataset (2019 + 2020).

〔evaluation〕
The accuracy rate (learning) was defined as the probability that the label of the estimated data is the same as the label of the nutritional state data corresponding to the position trajectory data of the input data. The accuracy rate (verification) was defined as the accuracy rate when data (verification data) different from the position trajectory data included in the training data was used as the input data. The evaluation results of the accuracy rate (learning) and accuracy rate (verification) for each dataset are shown in Table 1. The higher the accuracy rate (learning) and accuracy rate (verification), the higher the accuracy of the estimation by the nutritional state estimation system.

From the results shown in Table 1, it was found that a nutritional state estimation system having an estimation unit that estimates the nutritional state of livestock using output data related to the nutritional state of livestock obtained by inputting input data including position trajectory data representing the position trajectories of livestock into a trained model constructed by machine learning can easily estimate the nutritional state of livestock.

Example 2
In this example, a trained model was generated by machine learning the correlation between the position trajectory data of cattle and the health condition data of the cattle. The trained model was input with the position trajectory data, and output data was used to perform estimation, and the accuracy of the estimated data was evaluated.

[Generation of training data]
Five free-range cows (dairy cows) were classified by period, with five cows in a healthy period with few treatment histories and five cows in a diseased period with many treatment histories, classified into the first group and the second group, with five cows in a diseased period with many treatment histories, by referring to the data from the transmitter attached to each cow. The position trajectory data of the first group of cows was labeled as "good" and the position trajectory data of the second group of cows as "bad" as health condition data. This generated training data.

[Generating trained models, generating input data, obtaining estimation data, generating datasets]
The generation of the trained model, the generation of the input data, and the acquisition of the estimation data were performed in the same manner as in Example 1. The generation of the dataset was performed only once.

〔evaluation〕
The evaluation results of the accuracy rate (learning) and accuracy rate (verification) for the dataset are shown in Table 2. The higher the accuracy rate (learning) and accuracy rate (verification), the higher the accuracy of the estimation by the health condition estimation system.

From the results shown in Table 2, it was found that a health condition estimation system having an estimation unit that estimates the health condition of livestock using output data related to the health condition of livestock obtained by inputting input data including position trajectory data representing the position trajectories of livestock into a trained model constructed by machine learning can easily estimate the health condition of livestock.

This invention can be used in the livestock industry.

20 Nutritional state estimation device 24 Estimation unit 30 Data generation device 33 Data generation unit 40 Trained model generation device 43 Model generation unit 50 Feed amount determination device 53 Feed amount determination unit 100 Nutritional state estimation system 200 Feed amount determination system

Claims (8)

an estimation unit that estimates a state of a livestock by using, as input, position trajectory data representing the position trajectory of the livestock and status data representing the status of the livestock, using a trained model generated by performing machine learning using training data including position trajectory data representing the position trajectory of the livestock and status data representing the status of the livestock ;
The condition of the livestock is the nutritional state of the livestock or the health state of the livestock;
The position trajectory data includes image data representing the position trajectory of livestock.
State estimation system. The state estimation system according to claim 1 , further comprising a model generation unit that generates the trained model by performing machine learning using the position trajectory data and the state data as teacher data. 3. The state estimation system according to claim 1, further comprising a data generation unit that generates the position trajectory data by referring to position information of livestock that has been continuously acquired for a predetermined period of time. The method includes an estimation step of estimating a state of a livestock by using a trained model generated by performing machine learning using position trajectory data representing the position trajectory of the livestock and status data representing the status of the livestock as training data, and using the position trajectory data representing the position trajectory of the livestock as input;
The condition of the livestock is the nutritional state of the livestock or the health state of the livestock;
The position trajectory data includes image data representing the position trajectory of livestock.
State estimation methods. A feed amount determination system comprising a feed amount determination unit that determines an amount of feed to be fed to the livestock by referring to the state of the livestock estimated by the state estimation system according to any one of claims 1 to 3 . A feed amount determination method comprising a feed amount determination step of determining an amount of feed to be fed to the livestock by referring to the state of the livestock estimated by the state estimation method according to claim 4 . a model generation unit that performs machine learning using position trajectory data representing the position trajectory of livestock and status data representing the status of the livestock as training data to generate a trained model that estimates the status of the livestock using the position trajectory data representing the position trajectory of the livestock as input ;
The condition of the livestock is the nutritional state of the livestock or the health state of the livestock;
The position trajectory data includes image data representing the position trajectory of livestock.
Trained model generator. a model generation step of performing machine learning using position trajectory data representing the position trajectory of livestock and status data representing the status of the livestock as training data to generate a trained model for estimating the status of the livestock using the position trajectory data representing the position trajectory of the livestock as input;
The condition of the livestock is the nutritional state of the livestock or the health state of the livestock;
The position trajectory data includes image data representing the position trajectory of livestock.
How to generate a trained model.

Images