JP6764511B1 - Drone control device and pollen / fruiting agent automatic adhesion system using it, drone control program, drone control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a drone control device capable of automatically and highly accurately adhering pollen and fruit set agent to a pistil, an automatic pollen and fruit set agent adhering system using the drone control device, a drone control program, and a drone control method. An image data acquisition unit 42 that acquires image data of a flower from a drone 10, a flowering state determination unit 44 that determines whether or not a flower in the image data is blooming by a trained flowering determination model, and the like. The flower center identification unit 45 that identifies the pistil in the image data by the trained flower center identification model, the distance calculation unit 46 that calculates the distance from the pistil to the tip of the attachment attachment 14, and the tip of the attachment attachment 14 as the pistil. It has a control signal output unit 41 that outputs a control signal to be brought into contact with the drone. [Selection diagram] Fig. 1

Description

The present invention relates to a drone control device for automatically adhering pollen and fruit set agent to the pistil by a drone, an automatic pollen and fruit set agent adhering system using the drone control device, a drone control program, and a drone control method.

Conventionally, an artificial pollination technique for artificially attaching pollen to a pistil has been known for various plants such as fruit trees and vegetables. For example, Japanese Patent Application Laid-Open No. 2018-014929 states that artificial pollination is performed using a drone that can be remotely controlled, and the operator sprays pollen while controlling the timing and amount of pollen spraying by the remote controller. A pollinator is disclosed (Patent Document 1).

JP-A-2018-014929

However, in the invention described in Patent Document 1, manual operation of the drone by the operator is indispensable for the spraying work. Therefore, there is a problem that the more plants to be pollinated, the longer the time and labor required for the spraying work, and the greater the burden on the operator. In addition, since manual operation of the drone requires skill, there is also a problem that it is difficult to secure such human resources.

Further, in Patent Document 1, pollen is merely sprayed from above over the entire growing range of the plant to be pollinated. For this reason, pollen is less likely to adhere to the pistil, especially for flowers that bloom downward and flowers that bloom below the leaves, which may cause uneven fruit set and a decrease in yield. Further, in Patent Document 1, as the pollen to be sprayed, pollen for artificial pollination mixed with a bulking agent is used. For this reason, if pollen is sprayed over the entire area in the dark clouds, the amount of pollen sprayed increases unnecessarily, and the pollen cost increases.

The present invention has been made to solve such a problem, and is a drone control device capable of automatically and highly accurately adhering pollen and a fruit-setting agent to a pistil, and pollen using the drone control device. The purpose is to provide an automatic pollen adhesion system, a drone control program, and a drone control method.

The drone control device according to the present invention is a drone provided with an attachment for attaching pollen or fruit set agent to the pistil in order to solve the problem of automatically and highly accurately adhering pollen or fruit set agent to the pistil. It is a drone control device that controls the pistil, and is pre-learned using an image data acquisition unit that acquires pollen image data from the drone and teacher data consisting of a flower learning image and information on the flowering state of the flower. A flowering state determination unit that determines whether or not the flower imaged in the image data is blooming to the extent that pollen or a fruit-setting agent can be attached by the trained flowering determination model, and an image for learning the flower. The pistil included in the image data of the flower determined to be in bloom by the flowering state determination unit is identified by the trained flower heart identification model pre-trained using the teacher data consisting of the information about the pistil of the flower and the pistil of the flower. Based on the distance calculation unit that calculates the distance from the pistil specified by the pollen identification unit to the tip of the attachment, and the distance calculated by the distance calculation unit, the attachment attachment. It has a control signal to bring the tip of the pistil into contact with the pistil, or a control signal output unit to output a control signal to bring the tip of the attachment to the pistil and spray the pollen to the drone.

Further, as one aspect of the present invention, in order to solve the problem of automatically and highly accurately collecting fresh pollen from the stamens, the flower center specifying portion includes the image for learning the flower and information on the stamens of the flower. The stamens included in the image data of the flowers determined to be blooming by the flowering state determination unit are identified by the trained flower heart identification model pre-trained using the teacher data consisting of the above, and the distance calculation unit , The distance from the stamen specified by the pollen identification unit to the tip of the attachment attachment is calculated, and the control signal output unit calculates the tip of the attachment attachment based on the distance calculated by the distance calculation unit. May output to the drone a control signal that brings the stamen into contact with the stamen.

Further, as one aspect of the present invention, in order to solve the problem of determining with high accuracy whether or not pollen or a fruit-setting agent is flowering to the extent that it can be attached, the information on the flowering state is provided before flowering. It is composed of the shape and color of the flower according to the passage of time after flowering, and in the flowering state determination unit, the petals of the flower imaged in the image data are 60% or more open, and the petals and the center of the flower are It may be determined that the flower is in bloom when there is no wilting or discoloration.

Further, as one aspect of the present invention, in order to solve the problem of adhering pollen to the pistil of a flower blooming in any direction without leakage, the attachment attachment for adhering pollen to the pistil is the attachment of the flower to be pollinated. It is composed of an arm member that can be bent according to the orientation. When the flower is facing upward, the tip of the attachment attachment is directed downward, and when the flower is facing sideways, the tip of the attachment attachment is directed substantially horizontally. When the flower is facing downward, the tip of the attachment may be directed diagonally upward.

Further, as one aspect of the present invention, in order to solve the problem of adhering the fruit set agent to the pistil with high accuracy, the attachment attachment for adhering the fruit set agent to the pistil mists the fruit set agent by ultrasonic waves. It may be composed of an ultrasonic spray that is converted and sprayed.

The pollen / fruit set agent automatic attachment system according to the present invention is described in any one of claims 1 to 5 in order to solve the problem of automatically and highly accurately attaching pollen and fruit set agent to the pistil. The drone control device and the drone controlled by the drone control device.

Further, the drone control program according to the present invention is provided with an attachment for attaching pollen or fruiting agent to the pistil in order to solve the problem of automatically and accurately adhering pollen or fruiting agent to the pistil. It is a drone control program that controls a drone, and pre-learns using an image data acquisition unit that acquires flower image data from the drone, and teacher data consisting of a flower learning image and information on the flowering state of the flower. The flowering state determination unit that determines whether or not the flower imaged in the image data is blooming to the extent that pollen or a fruit-setting agent can be attached by the trained flowering determination model, and the learning of the flower. The pistil included in the image data of the flower determined to be in bloom by the flowering state determination unit by the trained flower center identification model pre-trained using the teacher data consisting of the image for use and the information on the pistil of the flower. Based on the distance calculation unit that calculates the distance from the pistil specified by the flower core identification unit to the tip of the attachment attachment, and the distance calculated by the distance calculation unit. A computer as a control signal output unit that outputs a control signal that brings the tip of the attachment to contact the pistil or a control signal that causes the tip of the attachment to be close to the pistil and sprays the fruiting agent to the drone. To work.

Further, the drone control method according to the present invention includes an attachment for attaching pollen or a fruit-setting agent to the pistil in order to solve the problem of automatically and accurately adhering pollen or a fruit-setting agent to the pistil. It is a drone control method for controlling a drone, and is pre-learned using an image data acquisition step of acquiring flower image data from the drone and teacher data consisting of a flower learning image and information on the flowering state of the flower. A flowering state determination step of determining whether or not the flower imaged in the image data is blooming to the extent that pollen or a fruit-setting agent can be attached by the trained flowering determination model, and learning of the flower. The pistil included in the image data of the flower determined to be in bloom in the flowering state determination step by the trained flower center identification model pre-trained using the teacher data consisting of the image for use and the information on the pistil of the flower. Based on the pollen identification step for specifying, the distance calculation step for calculating the distance from the pistil specified in the pollen identification step to the tip of the attachment for attachment, and the distance calculated in the distance calculation step, the attachment It has a control signal output step of outputting a control signal for bringing the tip of the attachment to contact the pistil, or a control signal for bringing the tip of the attachment close to the pistil and spraying the pollen, to the drone.

According to the present invention, pollen and a fruit-setting agent can be automatically and highly accurately attached to the pistil.

It is a block diagram which shows 1st Embodiment of the pollen / fruit-fruiting agent automatic adhesion system which comprises a drone control device and a drone which concerns on this invention. In the first embodiment, it is a figure which shows the use example of the attachment for attachment in the case of (a) the flower facing upward, (b) the flower facing sideways, and (c) the flower facing downward. It is a flowchart which shows the process of collecting pollen from a stamen and adhering it to a pistil by a drone control device of this 1st Embodiment, an automatic pollen / fruiting agent adhesion system using this, a drone control program, and a drone control method. In the second embodiment, it is a figure which shows the drone which attached the attachment for attachment which attaches a fruiting agent to a pistil. It is a flowchart which shows the process of adhering a fruit set agent to a pistil by a drone control device of this 2nd Embodiment, an automatic pollen / fruit-fruit adhering system using this, a drone control program, and a drone control method.

Hereinafter, a drone control device according to the present invention, an automatic pollen / fruit-fruit adhering system using the drone control device, a drone control program, and a first embodiment of a drone control method will be described with reference to the drawings.

As shown in FIG. 1, the pollen / fruit-fruiting agent automatic adhesion system 100 of the first embodiment includes a drone 10 which is a remotely controllable flying object and a drone control device 1 which remotely controls the drone 10 by a control signal. And have. Hereinafter, each configuration will be described in detail.

In the first embodiment, the drone 10 is composed of a multicopter type unmanned aerial vehicle equipped with a plurality of rotary wings, and mainly includes a camera 11 for capturing images of flowers and the like, and a state and an object of the drone 10. It has various sensors 12 for detecting the above, a controller 13 for performing various controls on the drone 10, and an attachment 14 for adhering pollen to the pistil.

As the camera 11, an RGB camera is provided at the lower part of the drone 10. Further, the camera 11 is controlled by the controller 13, and an image of a flower is taken at a predetermined timing, and the taken image is output to the controller 13. In the first embodiment, the camera 11 is provided at the lower part of the drone 10, but if it is possible to take an image of a flower, it may be provided at the side surface of the machine, and both the lower part and the side surface of the machine body may be provided. It may be provided in.

The sensor 12 includes an optical sensor that detects an object, a barometric pressure sensor that detects the altitude of the drone 10, a gyro sensor that detects the posture of the drone 10, and a GPS sensor that detects the current position of the drone 10. ing. In the first embodiment, the optical sensor employs a stereo vision method that calculates the distance to an object by the parallax of a pair of optical cameras, and is used at four locations on the front, back, left, and right sides of the drone 10. It is provided. However, the sensor for detecting an object is not limited to the stereo vision type, and may be an ultrasonic type, an infrared type, or a laser type.

The controller 13 is composed of a CPU (Central Processing Unit) and the like, and controls each configuration of the drone 10 according to a control signal from the drone control device 1. Further, the controller 13 acquires the output from each sensor 12 and transmits it to the drone control device 1, and controls the rotation speed of the rotor blades and the like so that the body of the drone 10 maintains an appropriate state.

The attachment attachment 14 is for adhering pollen to the pistil. In the first embodiment, as shown in FIG. 1, the attachment attachment 14 is composed of an arm member 14a that can be bent according to the direction of the flower to be pollinated. Further, a pollen holding portion 14b for holding pollen is provided at the tip of the attachment 14 for attachment.

In the first embodiment, the orientation of the attachment 14 is adjusted by adjusting the angle of the arm member 14a according to the flowering characteristics of the flower to be pollinated, as shown in FIG. Specifically, by setting as follows, the pollen holding portion 14b comes into contact with the pistil of a flower that blooms in any direction, and pollen can be attached without leakage.
When the flower is facing upward, the tip of the attachment 14 is directed downward.
When the flower is oriented sideways, the tip of the attachment 14 is directed substantially horizontally.
When the flower is facing downward, the tip of the attachment 14 is directed diagonally upward.

Next, the drone control device 1 of the first embodiment will be described.

The drone control device 1 is composed of a computer such as a personal computer or a smartphone, and controls the drone 10. In the first embodiment, as shown in FIG. 1, the drone control device 1 mainly includes a communication means 2 that performs wireless communication with the drone 10, the drone control program 1a of the first embodiment, and various types. It has a storage means 3 for storing data and a calculation processing means 4 for executing various calculation processes and functioning as each component described later. Hereinafter, each component means of the drone control device 1 will be described.

The communication means 2 implements a wireless communication function in the drone control device 1. In the first embodiment, the communication means 2 wirelessly connects to the drone 10 by using a wireless LAN standard such as Wi-Fi (registered trademark). Specifically, as shown in FIG. 1, a control signal is transmitted from the drone control device 1 to the drone 10, and captured image data or the like is transmitted from the drone 10 to the drone control device 1.

The storage means 3 is composed of a hard disk, a ROM (Read Only Memory), a RAM (Random Access Memory), a flash memory, and the like, and stores various data, and the arithmetic processing means 4 executes various arithmetic processes. It functions as a working area. In the first embodiment, as shown in FIG. 1, the storage means 3 includes a program storage unit 31, a flowering determination parameter storage unit 32, a flower center identification parameter storage unit 33, and a flight parameter storage unit 34. have.

The drone control program 1a of the first embodiment is installed in the program storage unit 31. Then, by executing the drone control program 1a by the arithmetic processing means 4, the computer functions as each component constituting the drone control device 1 of the first embodiment.

The usage pattern of the drone control program 1a is not limited to the above configuration. For example, the drone control program 1a may be stored in a non-temporary recording medium that can be read by a computer, such as a memory card, and read directly from the recording medium for execution. Further, it may be used by a cloud computing method or an ASP (Application Service Provider) method from an external server or the like.

The flowering determination parameter storage unit 32 stores the learned flowering determination parameters obtained by prior learning. In the first embodiment, as will be described later, as an algorithm for determining the flowering state, a learned flowering determination model obtained by applying a flowering determination parameter to a learning / inference model which is a machine learning algorithm is adopted. There is. Therefore, the flowering determination parameter storage unit 32 stores the flowering determination parameters obtained by pre-learning the flowering determination model using the teacher data consisting of the flower learning image and the information on the flowering state of the flower. Has been done.

The flower center specifying parameter storage unit 33 stores the learned flower center specifying parameters obtained by pre-learning. In the first embodiment, as described later, as an algorithm for identifying pistils and stamens, a trained flower center identification model obtained by applying a flower center identification parameter to a learning / inference model which is a machine learning algorithm is adopted. ing. Therefore, in the flower center identification parameter storage unit 33, the flower center identification parameter obtained by pre-learning the flower center identification model using the teacher data consisting of the flower learning image and the information on the pistil and stamen of the flower is obtained. It is remembered.

The flower learning image to be the teacher data may be directly taken by the user or may be taken by using the drone 10. Further, as the information on the flowering state of the flower, information on the shape and color of the flower according to the passage of time from before flowering to after flowering is used. Further, as information on the pistil and stamen of the flower, information on the shape and color of each of the pistil and stamen is used.

The flight parameter storage unit 34 stores the learned flight parameters obtained by the pre-learning. In the first embodiment, as the flight algorithm of the drone 10, a learned flight path determination model obtained by applying flight parameters to a learning / inference model which is a machine learning algorithm is adopted. For this reason, the flight parameter storage unit 34 has a general flight path from taking off from the initial position of the charging stand or the like to approaching the flower to be pollinated, pollinating it, returning to the initial position, and landing. Using image data, information on the drone 10 in each situation (distance to the object, altitude, acceleration, etc.), and teacher data consisting of position information on the flower to be pollinated and the initial position, it can be used as a flight path determination model. The flight parameters obtained by pre-learning are stored.

The arithmetic processing means 4 is composed of a CPU (Central Processing Unit) or the like, and by executing the drone control program 1a installed in the program storage unit 31, the control signal output unit 41 and the control signal output unit 41 are executed as shown in FIG. The image data acquisition unit 42, the flight path determination unit 43, the flowering state determination unit 44, the flower center identification unit 45, and the distance calculation unit 46 are configured to function. Hereinafter, each component will be described in more detail.

The control signal output unit 41 outputs a control signal for controlling the drone 10. In the first embodiment, the control signal output unit 41 controls to move a predetermined flight path to the drone 10 based on the determination result by the flight path determination unit 43, the output value of the sensor 12 received from the drone, and the like. Output a signal. As a predetermined flight path, a flight path is set in advance from taking off from the initial position of the charging stand or the like, approaching the flower to be pollinated, pollinating the flower, returning to the initial position, and landing.

Further, in the first embodiment, the control signal output unit 41 is provided at the tip of the attachment attachment 14 based on the distance from the pistil or stamen calculated by the distance calculation unit 46 to the tip of the attachment attachment 14. A control signal for bringing the pollen holding portion 14b into contact with the pistil or stamen is output. It should be noted that the pollen collection rate from the stamens and the pollen adhesion rate to the pistils may be increased by vibrating the tip of the attachment for attachment in addition to the contact.

The image data acquisition unit 42 acquires the image data taken by the drone 10. In the first embodiment, the image data acquisition unit 42 periodically acquires image data at short time intervals while the drone 10 is flying on the flight path, and the flight path determination unit 43 blooms as needed. It is provided to the state determination unit 44 and the flower center identification unit 45.

The flight path determination unit 43 determines the flight path of the drone 10. In the first embodiment, the flight path determination unit 43 is composed of a learned flight path determination model obtained by applying flight parameters to a learning / inference model which is a machine learning algorithm. Then, when determining the flight path, the flight path determination unit 43 applies the learned flight parameters stored in the flight parameter storage unit 34 to the flight path determination model, and determines the learned flight route. The image data acquired by the image data acquisition unit 42 is input to the model. As a result, it is determined whether or not the aircraft is flying in a preset flight path without colliding with the object captured in the image data.

The flowering state determination unit 44 determines the flowering state of the flower captured in the image data. In the first embodiment, the flowering state determination unit 44 is composed of a learned flowering determination model obtained by applying a flowering determination parameter to a learning / inference model which is a machine learning algorithm. Then, when determining the flowering state of the flower, the flowering state determination unit 44 applies the learned flowering determination parameters stored in the flowering determination parameter storage unit 32 to the flowering determination model, and the learned parameters are applied to the flowering determination model. The flower image data acquired by the image data acquisition unit 42 is input to the flowering determination model. Thereby, it is determined whether or not the flower captured in the image data is blooming to the extent that pollen can be attached.

In the first embodiment, the flowering state determination unit 44 states that the flower is blooming when the petals of the flower captured in the image data are 60% or more open and the petals and the corolla are not wilted or discolored. judge. However, the criteria for determining flowering are not limited to the above, and may be appropriately changed depending on the characteristics of the plant to be pollinated.

The flower center specifying unit 45 identifies the flower center of the flower captured in the image data. In the first embodiment, the flower center specifying unit 45 is configured by a learned flower center specifying model obtained by applying a flower center specifying parameter to a learning / inference model which is a machine learning algorithm. Then, when specifying the flower heart of the flower, the flower center specifying unit 45 applies the learned flower center specifying parameter stored in the flower center specifying parameter storage unit 33 to the flower center specifying model, and identifies the learned flower center. The image data of the flower acquired by the image data acquisition unit 42 is input to the model. In this way, the pistil and stamen of the flower captured in the image data are identified.

In the first embodiment, a convolutional neural network (CNN), which is a model specialized for image classification among deep learning methods, is adopted as a machine learning algorithm. Other machine learning algorithms may be used without limitation.

The distance calculation unit 46 calculates the distance from the pistil or stamen to the tip of the attachment 14. In the first embodiment, the distance calculation unit 46 first determines the distance from the pistil or stamen specified by the flower center specifying unit 45 to the optical sensor 12 based on the output value from the optical sensor 12 provided on the drone 10. Is calculated. Then, the distance calculation unit 46 calculates the distance from the pistil or stamen to the tip of the attachment 14 based on the difference information between the optical sensor 12 and the tip of the attachment 14 registered in advance. It has become.

Next, the actions of the drone control device 1 of the first embodiment, the pollen / fruit fruit automatic adhesion system 100 using the drone control device 1, the drone control program 1a, and the drone control method will be described. In the first embodiment, a process in which fresh pollen is collected from the stamen by the attachment attachment 14 and then the pollen is attached to the pistil will be described.

First, as shown in FIG. 3, the control signal output unit 41 outputs a control signal for moving to the flower to be pollinated to the drone 10 (step S1). As a result, the drone 10 takes off from the initial position of the charging stand or the like and starts flying according to a predetermined flight path. While the drone 10 is flying, the image data acquisition unit 42 periodically acquires image data (step S2). As a result, the flight path determination unit 43 constantly determines whether or not the drone 10 is flying on a preset flight path without colliding with an object.

When the drone 10 arrives near the target flower, the flowering state determination unit 44 determines the flowering state of the flower captured in the image data (step S3). As a result of the determination, when the pollen has not bloomed to the extent that it can be attached (step S3: NO), the pollination treatment for the flower is stopped, and the process proceeds to step S10 described later. On the other hand, when the flower is blooming to the extent that pollen can be attached (step S3: YES), the flower center specifying portion 45 searches for a stamen in the flower (step S4).

At this time, in the first embodiment, when determining whether or not a flower is blooming, it is used as a criterion for determining flowering that the petals of the flower are 60% or more open and that the petals and the heart of the flower are not wilted or discolored. ing. Therefore, it is determined with high accuracy whether or not the flowering is such that pollen can be attached.

When the stamen is specified by the flower center specifying unit 45 (step S4: YES), the distance calculation unit 46 calculates the distance from the stamen to the tip of the attachment 14 (step S5). Then, based on the distance, the control signal output unit 41 outputs a control signal that brings the tip of the attachment 14 into contact with the stamen (step S6). As a result, the tip of the attachment 14 adheres to the stamen, so that fresh pollen produced by the stamen is automatically and highly accurately collected.

When pollen is collected from the stamens, the flower core identification unit 45 separately searches for the pistil (step S7). Then, when the pistil is specified by the flower center specifying unit 45 (step S7: YES), the distance calculation unit 46 calculates the distance from the pistil to the tip of the attachment attachment 14 (step S8). Then, based on the distance, the control signal output unit 41 outputs a control signal that brings the tip of the attachment 14 into contact with the pistil (step S9). As a result, the tip of the attachment 14 adheres to the pistil, so that fresh pollen collected from the stamen is automatically and highly accurately adhered to the pistil, and the pollination work is completed.

At this time, in the first embodiment, the attachment attachment 14 is set at an angle according to the direction of the flower to be pollinated. Therefore, pollen is collected from the stamens of flowers that bloom in any direction without leakage, and pollen is attached to the pistils without leakage.

In general, the freshness of pollen after flowering is important, and it is not preferable to store and use pollen. Therefore, in the first embodiment, fresh pollen is secured by automating the operation of collecting pollen from the stamens. However, for plants in which a sufficient fruit set rate can be expected even with pollen for artificial pollination, it is possible to attach pollen to the tip of the attachment 14 in advance and perform only the operation of attaching the pollen to the pistil. Good. In this case, the processes from step S4 to step S6 are unnecessary.

After the pollination work is completed, the drone 10 has a remaining battery level (step S10: YES), and when pollinating another flower (step S11: YES), the process returns to step S1 and the above-mentioned process is repeated. On the other hand, when the remaining battery level of the drone 10 is low (step S10: NO) or when other flowers are not pollinated (step S11: NO), the flight path determination unit 43 calculates the distance from the current position to the initial position. (Step S12), a control signal for returning to the initial position is output to the drone 10 (step S13).

According to the first embodiment as described above, the following effects are obtained.
1. 1. Pollen and fruit-setting agents can be attached to the pistil automatically and with high accuracy.
2. 2. Fresh pollen can be collected automatically and with high accuracy from the stamens, and the fruit set rate can be improved.
3. 3. It is possible to determine with high accuracy whether or not the flowers are blooming to the extent that pollen can be attached.
4. Pollen can be attached to the pistil of any flower that blooms in any direction.
5. By repeating machine learning, it is finally possible to completely automate the work of attaching pollen and fruit set agents.

Next, the drone control device 1 according to the present invention, the pollen / fruit set agent automatic adhesion system 100 using the drone control device 1, the drone control program 1a, and the second embodiment of the drone control method will be described. Of the configurations of the second embodiment, the same or equivalent configurations as those of the first embodiment described above are designated by the same reference numerals, and the description thereof will be omitted again.

The features of the drone control device 1 of the second embodiment, the pollen / fruiting agent automatic adhesion system 100 using the drone control device 1, the drone control program 1a, and the drone control method are the adhesion of pollen to the pistil in the first embodiment. Instead of the attachment 14, the drone is equipped with an attachment 15 for attaching a fruit-setting agent to the pistil.

In the second embodiment, the attachment attachment 15 for adhering the fruit set agent to the pistil is configured by an ultrasonic spray that atomizes and sprays the fruit set agent by ultrasonic waves, as shown in FIG. Then, depending on the growing condition of the plant, the attachment attachment 14 for attachment 14 of the first embodiment and the attachment 15 for attachment of the second embodiment are replaced, and the drone 10 is used in both the pollination work and the attachment work of the fruit set agent. It can be reused.

Next, the actions of the drone control device 1 of the second embodiment, the pollen / fruit fruit automatic adhesion system 100 using the drone control device 1, the drone control program 1a, and the drone control method will be described. Of the steps of the second embodiment, the same or corresponding steps as those of the first embodiment described above are designated by the same reference numerals, and the description thereof will be omitted again.

As shown in FIG. 5, in the second embodiment, first, the processes from step S1 to step S3 are executed to search for flowers that are blooming to the extent that a fruit-setting agent can be attached. At this time, the flowering state determination unit 44 determines the flowering state using the same determination criteria as in the first embodiment.

Next, the processes of steps S7 and S8 are executed, and the distance from the pistil of the flowering flower to the tip of the attachment 15 for attachment is calculated. Then, based on the distance, the control signal output unit 41 outputs a control signal to the drone 10 so that the tip of the attachment 15 for attachment is brought close to the pistil and the fruit fruit is injected (step S21). As a result, the fruit set agent is sprayed from the tip of the attachment attachment 15, so that the fruit set agent is automatically and highly accurately attached to the pistil.

After that, the processing of steps S10 and S11 determines whether or not to inject the fruit set agent onto other flowers, and if not, the drone is returned to the initial position by the processing of steps S12 and S13. Let me.

According to the second embodiment as described above, in addition to the action and effect of the first embodiment, it is possible to automatically and highly accurately attach the fruit-setting agent to the pistil.

The present invention is not limited to the above-described embodiments, and can be appropriately modified. For example, in each of the above-described embodiments, the remaining battery level of the drone 10 is confirmed when the pollination work and the spraying work are continued (step S10). However, the present invention is not limited to this configuration, and even when taking off from the initial position, the remaining battery level may be checked and the takeoff may be performed only when the battery is sufficiently charged.

Further, in each of the above-described embodiments, the flowering determination parameter, the flower center identification parameter, and the flight parameter that have been learned in advance are prepared, but each time the drone 10 is flown, image data or the like is acquired and machine learning is performed. It may be made to update each learned parameter sequentially.

Further, in the above-described first embodiment, the orientation of the attachment attachment 14 is manually adjusted, but the present invention is not limited to this, and the orientation may be automatically adjusted. Specifically, in the drone 10, the bending angle of the arm member 14a can be driven and controlled. On the other hand, the drone control device 1 determines the type and orientation of the flower by image recognition from the image data, and outputs a control signal for designating the bending angle according to the determination result.

Further, when collecting teacher data by the drone 10, the drone 10 may be manually operated by a smartphone, a tablet or the like.

1 Drone control device 1a Drone control program 2 Communication means 3 Storage means 4 Arithmetic processing means 10 Drone 11 Camera 12 Sensor 13 Controller 14 Attachment for attachment (1st embodiment)
14a Arm member 14b Pollen holding part 15 Attachment for attachment (second embodiment)
31 Program storage unit 32 Flowering determination parameter storage unit 33 Flower center identification parameter storage unit 34 Flight parameter storage unit 41 Control signal output unit 42 Image data acquisition unit 43 Flight path determination unit 44 Flowering state determination unit 45 Flower core identification unit 46 Distance Calculation unit 100 Automatic adhesion system for pollen and fruit set

Claims (7)

A drone control device that controls a drone equipped with an attachment for attaching pollen or fruit set agent to the pistil.
An image data acquisition unit that acquires flower image data from the drone,
The flowering judgment model of learned obtained by prior learning using teacher data composed of information relating to flowering state of the learning image and Floral, the image data being captured in flowers enough to be attached to pollen A flowering state determination unit that determines whether or not it is in bloom,
A flower determined to be flowering by the flowering state determination unit by a trained flower center identification model pre-trained using teacher data consisting of a flower learning image and information on flower pistils and stamens . The flower heart identification part that identifies the pistil and stamen included in the image data,
A distance calculation unit that calculates the distance from the pistil and stamen specified by the flower center identification unit to the tip of the attachment attachment.
Based on the distance calculated by the distance calculation unit, a control signal is brought into contact with the tip of the attachment for attachment to the pistil, and control for outputting the tip of the attachment for attachment to the drone control signals of contacting the stamen Signal output section and
Have a,
For a flower determined to be in bloom by the flowering state determination unit
The pollen identification unit identifies the stamen, the distance calculation unit calculates the distance from the stamen to the tip of the attachment, and the control signal output unit determines the tip of the attachment attachment based on the distance. After collecting pollen from the stamen by outputting a control signal for contacting the stamen,
The pollen identification unit identifies the pistil, the distance calculation unit calculates the distance from the pistil to the tip of the attachment, and based on the distance, the control signal output unit determines the tip of the attachment. A drone control device that attaches pollen collected from the pistil to the pistil by outputting a control signal for contacting the pistil . A set of image data in the flight path from when the drone took off from the initial position to approaching the flower to be polluted, polluting, returning to the initial position and landing, and information on the drone in each situation. The drone collides with an object imaged in the image data by a learned flight path determination model pre-trained using teacher data consisting of a flower to be polluted and position information of the initial position. The drone control device according to claim 1, further comprising a flight path determination unit that determines whether or not the aircraft is flying on a preset flight path. The information on the flowering state is composed of the shape and color of the flower according to the passage of time from before flowering to after flowering.
The flowering state determination unit determines that the flower is blooming when the petals of the flower captured in the image data are 60% or more open and the petals and the flower heart are not wilted or discolored. Alternatively, the drone control device according to claim 2. The attachment attachment for attaching pollen to the pistil is composed of an arm member that can be bent according to the direction of the flower to be pollinated.
If the flower is facing up, point the tip of the attachment to the bottom.
If the flower is oriented sideways, point the tip of the attachment attachment approximately horizontally.
The drone control device according to any one of claims 1 to 3, wherein when the flower is facing downward, the tip of the attachment attachment is directed obliquely upward. An automatic pollen / fruit-fruit adhering system comprising the drone control device according to any one of claims 1 to 4 and a drone controlled by the drone control device. A drone control program that controls a drone with an attachment that attaches pollen or fruit set to the pistil.
An image data acquisition unit that acquires flower image data from the drone,
The flowering judgment model of learned obtained by prior learning using teacher data composed of information relating to flowering state of the learning image and Floral, the image data being captured in flowers enough to be attached to pollen A flowering state determination unit that determines whether or not it is in bloom,
A flower determined to be flowering by the flowering state determination unit by a trained flower center identification model pre-trained using teacher data consisting of a flower learning image and information on flower pistils and stamens . The flower heart identification part that identifies the pistil and stamen included in the image data,
A distance calculation unit that calculates the distance from the pistil and stamen specified by the flower center identification unit to the tip of the attachment attachment.
Based on the distance calculated by the distance calculation unit, a control signal is brought into contact with the tip of the attachment for attachment to the pistil, and control for outputting the tip of the attachment for attachment to the drone control signals of contacting the stamen Signal output section and
With the functioning of the computer and,
For a flower determined to be in bloom by the flowering state determination unit
The pollen identification unit identifies the stamen, the distance calculation unit calculates the distance from the stamen to the tip of the attachment, and the control signal output unit determines the tip of the attachment attachment based on the distance. After collecting pollen from the stamen by outputting a control signal for contacting the stamen,
The flower center specifying unit identifies the pistil, the distance calculating unit calculates the distance from the pistil to the tip of the attachment, and based on the distance, the control signal output unit determines the tip of the attachment. A drone control program that attaches pollen collected from the pistil to the pistil by outputting a control signal for contacting the pistil . A drone control method that controls a drone equipped with an attachment for attaching pollen or fruiting agent to the pistil.
An image data acquisition step for acquiring flower image data from the drone, and
The flowering judgment model of learned obtained by prior learning using teacher data composed of information relating to flowering state of the learning image and Floral, the image data being captured in flowers enough to be attached to pollen A flowering state determination step to determine whether or not it is in bloom,
A flower determined to be in bloom in the flowering state determination step by a trained flower center identification model pre-trained using teacher data consisting of an image for learning the flower and information on the pistil and stamen of the flower. A flower heart identification step to identify the pistil and stamen included in the image data,
A distance calculation step for calculating the distance from the pistil and stamen specified in the flower center specifying step to the tip of the attachment attachment, and
Based on the distance calculated by said distance calculation step, the control signals contacting tip of the attachment for attachment to the pistil, and control for outputting the tip of the attachment for attachment to the drone control signals of contacting the stamen Signal output step and
Have a,
For a flower determined to be in bloom in the flowering state determination step
The stamen is specified in the pollen identification step, the distance from the stamen to the tip of the attachment is calculated in the distance calculation step, and the tip of the attachment is set in the control signal output step based on the distance. After collecting pollen from the stamen by outputting a control signal for contacting the stamen,
The pistil is specified in the pollen identification step, the distance from the pistil to the tip of the attachment is calculated in the distance calculation step, and the tip of the attachment is set in the control signal output step based on the distance. A drone control method in which pollen collected from the pistil is attached to the pistil by outputting a control signal for contacting the pistil .

Images