Bird and animal repellent system, bird and animal repellent method and program

Abstract

To provide a bird and animal repellent system capable of further quickly identifying repellent parameters with high repellent effects compared to conventional methods and having long-lasting repellent effects.SOLUTION: A bird and animal repellent system includes a repellent effect calculation unit 15 and an optimization processing unit 16. The repellent effect calculation unit 15 calculates a repellent effect y of repelling action performed during a sampling period SPi. The optimization processing unit 16 adds implementation data 72 consisting of a recommended repellent parameter 75 used in the repelling action and the calculated repellent effect y as new regression data to a regression data storage unit 18, then executes Bayesian optimization processing on the basis of an implementation data group stored in the regression data storage unit 18, derives the repelling parameter Xnext corresponding to a point where an acquisition function is maximized, and sets it as the recommended repellent parameter 75. In a next sampling period SPi+1, the repelling action is performed on the basis of a newly set recommended repellent parameter 75 (Xi+1).SELECTED DRAWING: Figure 12

Description

[Technical Field]

[0001] The present invention relates to a bird and animal repelling system, a bird and animal repelling method, and a program for repelling birds and animals that have invaded a monitored area. [Background technology]

[0002] Recently, there has been an increase in cases where birds and animals have invaded managed fields, causing economic and human damage. For example, in grape farms, a problem has become apparent where birds fly in in large numbers and eat the harvested crops in the short 1-2 months between the time the grapes ripen and the end of the harvest.

[0003] It is said that when birds and animals experience something out of the ordinary (a stimulus), they become disgusted and run away from the area, and will be cautious and not approach the area again for a while. Based on this characteristic, efforts have been made to scare away birds and animals by using "repellent means" such as flying objects, sounds, lights, and projectiles to provide stimuli that repel birds and animals. However, it has also been pointed out that the stimuli provided by the repellent means do not necessarily cause any actual harm to birds and animals, and as the number of times the stimuli are provided increases, the birds and animals become accustomed to the stimuli, ultimately resulting in the repellent effect not lasting (see Chapters 1-5 and 2-1 of Non-Patent Document 1; also see Non-Patent Document 2).

[0004] For this reason, some countermeasures have been proposed, such as preparing multiple repellent measures in advance and switching between them when the repellent effect begins to decrease, or combining multiple repellent measures simultaneously while sequentially changing the "combination pattern" (see Chapters 2-3C of Non-Patent Document 1 and Example 3 of Patent Document 1). Note that the "combination pattern of repellent measures" does not only refer to the combination of repellent measures themselves, but also includes variations due to differences in the form and degree of irritation of each repellent measure combined. In this specification, this is sometimes referred to as the "repellent parameter."

[0005] Machine learning is also used to select repellent parameters suitable for target birds and animals. For example, Patent Document 2 proposes generating an estimation model (22) by machine learning (paragraphs

[0022] ,

[0040] , etc.), and using this estimation model (22) to estimate the content of control of firing to have a more deterrent effect (paragraphs

[0021] ,

[0023] , etc.). [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] National Agriculture and Food Research Organization, "Bird Ecology and Damage Control - Focusing on Crows and Brown-eared Bulbuls," June 17, 2020 [Non-patent document 2] Kazuki Kobayashi and six others, "Development of a bird scare system based on bird behavior using deep learning," 2019 National Conference of the Japanese Society for Artificial Intelligence (33rd), Japanese Society for Artificial Intelligence, June 4, 2019, manuscript number IF'S-OS-Ai-02 [Patent documents]

[0007] [Patent Document 1] Patent Publication No. 2021-40519 [Patent Document 2] Patent Publication No. 2021-97610 Summary of the Invention [Problem to be solved by the invention]

[0008] However, in order to generate an estimation model using machine learning, in the case of the grape farm mentioned above, for example, a considerable number of days and periods are required, as learning is repeated on-site while waiting for birds to arrive. Therefore, before an estimation model adapted to the birds that have arrived can be established, the necessary control period of approximately one to two months will have passed, and bird repelling measures will not be able to be implemented in time, posing a risk of economic damage.

[0009] Incidentally, the optimal repellent parameters that maximize effectiveness are likely to change from time to time due to various changes in conditions, such as changes in the degree to which birds and animals become accustomed to the repellent, changes in the environment of the monitored area (the grape farm in the above example), and the appearance of birds and animals with different personalities.

[0010] On the other hand, the estimation model (22) described in Patent Document 2 is based on the results of past machine learning and is therefore fixed. Therefore, the estimation model (22) described in Patent Document 2 cannot respond to various changes in conditions that may occur in the future from the current point in time when birds and animals are being repelled, and cannot propose the most effective repellent parameters at any given time. Ultimately, the repellent effect of the technology described in Patent Document 2 does not last.

[0011] The present invention has been made in consideration of the above circumstances, and aims to provide a bird and animal repellent system that can find a highly effective combination pattern of repellent means (repellent parameters) more quickly than conventional methods and that has a long-lasting repellent effect. It also aims to provide such a bird and animal repellent method and a corresponding program. [Means for solving the problem]

[0012] [1] According to one aspect of the present invention, a bird and animal repelling system is provided that updates an "avoidance parameter" that quantitatively indicates the type of stimulus that repels birds and animals at each sampling period, and repels birds and animals in accordance with the avoidance parameter. This bird and animal repelling system includes at least a repelling controller that controls the repelling of birds and animals, and a repelling device that performs repelling actions against birds and animals in response to commands from the repelling controller. The repelling controller includes the following elements: A repelling command unit that, when it is determined that a bird or animal that needs to be repelled is staying within the monitoring area, causes the repelling device to carry out repelling action based on the location information of the bird or animal staying within the monitoring area and the recommended repelling parameters to be used during the sampling period. A repulsion effect calculation unit that calculates the repulsion effect y of the repulsion action performed during the sampling period when the sampling period ends The implementation data (x, y) consisting of the recommended repellent parameter x used in the repelling behavior during the sampling period and the repellent effect y calculated by the repellent effect calculation unit is added to the regression data storage unit as new regression data, and then Bayesian optimization is performed based on the implementation data group stored in the regression data storage unit, and the repellent parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized (maximum effect point) is calculated. next and derive the new repellent parameter x next The optimization section sets (substitutes) as the recommended avoidance parameter.

[0013] The repelling controller is configured so that, in the next sampling period, the repelling command unit causes the repelling device to perform repelling action based on the recommended repelling parameters newly set by the optimization processing unit. The repulsion controller is preferably equipped with a regression data storage unit that adds and stores regression data. Also, it is preferable that the repulsion controller is provided with an edge device (which can also be called a detection information transmission unit) outside the controller that transmits location information of a bird or animal that needs to be repelled to the repulsion controller when it is determined that the bird or animal is staying within the monitoring area.

[0014] [2] According to another aspect of the present invention, there is provided a bird and animal repelling method that updates an "avoidance parameter" that quantitatively indicates the type of stimulus that repels birds and animals at each sampling period, and repels birds and animals in accordance with the avoidance parameter. This bird and animal repelling method includes a repelling command step, a repelling effect calculation step, and an optimization processing step. The repelling command step, when it is determined that a bird or animal that needs to be repelled is staying within a monitoring area, causes the repelling device to carry out repelling action based on the location information of the staying bird or animal and the recommended repelling parameters to be used during the sampling period. The repelling effect calculation step, when the sampling period ends, calculates the repelling effect y of the repelling action carried out during the sampling period. The optimization processing step adds implementation data (x, y) consisting of the recommended repelling parameter x used in the repelling action during the sampling period and the repelling effect y calculated in the repelling effect calculation step to a regression data storage unit as new regression data, and then executes Bayesian optimization based on the implementation data group stored in the regression data storage unit, and calculates the repelling parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized (maximum effect point). next and derive the new repellent parameter x next is set as the recommended repelling parameter. Then, in the repelling command step in the next sampling period, the optimization processing step is performed to make the repelling device perform repelling behavior based on the newly set recommended repelling parameter.

[0015] [3] According to yet another aspect of the present invention, there is provided a program for use in a bird and animal repelling system which includes at least a repelling controller that controls the repelling of birds and animals and a repelling device that performs repelling actions against birds and animals to be repelled in response to commands from the repelling controller, and which updates ``repelling parameters'' that quantitatively indicate the type of stimuli that repel birds and animals at each sampling period and repels birds and animals in accordance with the repelling parameters. This program is configured to cause the computer constituting the chase controller to perform the following processes. That is, when it is determined that a bird or animal that needs to be repelled is staying within the monitoring area, a repelling command step causes the repelling device to perform a repelling action based on the position information of the staying bird or animal and the recommended repelling parameters to be used during the sampling period; a repelling effect calculation step, when the sampling period ends, calculates a repelling effect y by the repelling action performed during the sampling period; and, after adding the implementation data (x, y) consisting of the recommended repelling parameter x used in the repelling action during the sampling period and the repelling effect y calculated in the repelling effect calculation step to a regression data accumulation unit as new regression data, executes Bayesian optimization processing based on the implementation data group stored in the regression data accumulation unit, and calculates the repelling parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized (maximum effect point). next and derive the new repellent parameter x next as recommended repelling parameters, and in the repelling command step in the next sampling period, the repelling device is made to take repelling action based on the recommended repelling parameters newly set by performing the optimization processing step. [Effects of the Invention]

[0016] According to the present invention, it is possible to find a combination pattern (avoidance parameters) of repellent means with a high repellent effect more quickly than before, and to provide a bird and animal repellent system with a highly durable repellent effect. It is also possible to provide such a bird and animal repellent method and a corresponding program. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a system configuration diagram showing an overview of a bird and animal repellent system 1 according to a first embodiment. [Figure 2] FIG. 10 is a diagram for explaining an example of application of the avoidance parameter x. [Figure 3] 1 is a diagram illustrating an example of a hardware configuration of an information processing device 100. FIG. [Figure 4]1 is a diagram illustrating an example of a hardware configuration of a bird and animal repellent system 1 according to a first embodiment. [Figure 5] FIG. 2 is a functional block diagram showing the functional configuration of the edge device 30 of the first embodiment. [Figure 6] 10 is a sequence diagram illustrating a process from acquisition of image data 60 by the edge device 30 to step S22 in which the repelling controller 10 issues a command to repel birds and animals. [Figure 7] 1 is a diagram showing an example of the data structure of bird and animal stopping information 62 in the first embodiment. FIG. [Figure 8] 1 is a functional block diagram showing the functional configuration of a chase controller 10 of the first embodiment. [Figure 9] FIG. 10 is a diagram showing an example of data passed from the chase controller 10 to the drone 21 when a repelling command is issued in the first embodiment. [Figure 10] FIG. 10 is a sequence diagram mainly explaining the overall repelling action by the repelling device 20. [Figure 11] 10 is a sequence diagram mainly explaining a chase effect calculation step S30 and an optimization processing step S40 by the chase controller 10. FIG. [Figure 12] 1 is a diagram for explaining an example of the bird and animal repelling system 1 and bird and animal repelling method of embodiment 1. FIG. [Figure 13] FIG. 2 is a diagram for explaining data used in the optimization processing unit 16 of the first embodiment. [Figure 14] 10 is a schematic distribution map for explaining how the optimization processing unit 16 of the first embodiment derives an avoidance parameter xnext corresponding to a maximum effect point MEP. DETAILED DESCRIPTION OF THE INVENTION

[0018] The bird and animal repelling system, bird and animal repelling method, and program according to the present invention will be described below with reference to the drawings. Note that the explanations of symbols common to each drawing will be omitted in other drawings, as the contents already explained for those symbols can be used in the explanations of the other drawings. Furthermore, in later drawings, symbols may be omitted for parts that have already been indicated and explained by symbols in the previous drawings. The subscripts i and j used in the symbols are index numbers consisting of integers equal to or greater than 0.

[0019] [Embodiment 1] 1. Configuration of the bird and animal repellent system 1 according to the first embodiment FIG. 1 is a system configuration diagram showing an overview of a bird and animal repellent system 1 according to a first embodiment. FIG. 2 is a diagram for explaining an application example of a repellent parameter x. FIG. 3 is a diagram showing an example of the hardware configuration of an information processing device 100. Note that in this specification, for ease of understanding, "birds and animals" may be simply referred to as "birds" in the explanation. Conversely, the content can also be understood by reading the word "bird" as "bird and animal."

[0020] (1) Overview of the Bird and Animal Repellent System 1 1, the bird and animal repelling system 1 according to the first embodiment includes at least a repelling controller 10 that controls the repelling of birds and animals, and a repelling device 20 that performs repelling actions against birds and animals to be repelled in response to commands from the repelling controller 10. It is preferable that the system further includes an edge device 30 equipped with a camera 32a for monitoring whether birds and animals that need to be repelled are staying. The edge device 30 includes devices known as edge computing devices and so-called IoT (Internet of Things).

[0021] In the illustrated example, five edge devices 30 are installed with their cameras 32a directed toward the monitoring areas MA1 to MA5. -1 ~30 -5are placed, and three birds BD1 to BD3 that need to be chased away are depicted flying in and staying within the monitoring area MA2 at the same time. A communication path is established between the edge device 30 and the chase controller 10 via wireless communication means. In addition, a communication path is also established between the chase controller 10 and the drone 21 via a separate wireless communication means. An overview will be given below with reference to FIG. 1.

[0022] Edge Device 30 -2 When the edge device 30 captures an image of an area including the monitoring area MA2 as shown in the lower left of the figure and acquires image data, -2 The edge device 30 detects target birds and animals TBD1 to TBD3 using the image data. -2 reports the detection of the target birds and animals TBD1 to TBD3 together with the position data of the target birds and animals TBD1 to TBD3 to the repelling controller 10 via wireless communication means. The repelling controller 10 issues a command to repel the target birds and animals TBD1 to TBD3 to the repelling device 20 such as a drone 21 via wireless communication means, while passing the position information and repellent parameters of the target birds and animals TBD1 to TBD3.

[0023] As shown in Figure 2, "repel parameters" are parameters that quantitatively indicate the manner in which various stimuli (approaching objects such as drones, light irradiation, sound emission, scent emission, liquid spray, etc.) are applied that are believed to repel target birds and animals. In the example shown in the figure, the repel parameters X are shown as drone approach speed x1, drone flight altitude x2, wavelength of irradiated LED light x3, intensity of irradiated LED light x4, frequency of repellent sound x5, and volume of repellent sound x6, but are not limited to these and can be variously assumed. Upon receiving a command to repel the target birds and animals TBD1 to TBD3, the drone 21 etc. performs repelling action toward the location indicated by the location information while appropriately changing and setting the flight speed, flight altitude, light irradiation mode, sound emission mode, etc. based on the repelling parameter X.

[0024] The bird and animal repelling system 1 according to the first embodiment is a system that updates the "avoidance parameter x" that quantitatively indicates the type of stimulus that repels birds and animals for each sampling period SP, and repels birds and animals in accordance with the avoidance parameter x. Details of the system will be explained below.

[0025] (2) Hardware Configuration of Information Processing Device 100 The electronic devices such as the chase controller 10, the edge device 30, and the on-board computer 22 (described later) are realized using an information processing device 100.

[0026] 3, the information processing device 100 includes a processor 110, a memory 120, a storage 130, an input / output I / F (Interface) 140, and a communication I / F (Interface) 150. These are connected to a bus BS. The memory 120 and the storage 130 are collectively referred to as a storage unit 160.

[0027] The processor 110 operates based on a program stored in the storage unit 160 and controls each unit (functions shown in the functional block diagrams described below) described below. The memory 120 can be configured with volatile or non-volatile memory. The storage 130 can be configured with an auxiliary storage device such as an SDD (Solid State Drive) or HDD (Hard Disk Drive).

[0028] The communication I / F 150 is an interface that communicates with external devices. The communication I / F 150 receives data from external devices via a communication path or a network and sends the data to the processor 110, and transmits data generated by the processor 110 to external devices via the communication path or the network.

[0029] The input / output I / F 140 interfaces with input / output devices such as a camera, etc. The processor 110 acquires various data from the input / output devices via the input / output I / F 140 and controls the input / output devices via the input / output I / F 140.

[0030] (3) Example of hardware configuration for bird and animal repelling system 1 4 is a diagram showing an example of the hardware configuration of the bird and animal repelling system 1 according to embodiment 1. In the example shown in the figure, the bird and animal repelling system 1 is broadly divided into a repelling controller 10, a drone station 28 and a drone 21 constituting a repelling device 20, and an edge device 30. Of these, at least the on-board computer 22 constituting the edge device 30, the repelling controller 10, and the drone 21 incorporates the information processing devices 100A, 100B, and 100C described above.

[0031] The edge device 30 has a camera 32a that constitutes the imaging unit 32. The camera 32a is connected to, for example, an input / output I / F 140A of the information processing device 100A. The communication I / F 150A here has a function for performing low-power long-distance wireless communication compliant with the LoRa (registered trademark) standard. The memory unit 160A stores a trained model for identifying target birds and animals and programs for implementing various functions described below.

[0032] The communication I / F 150B1 of the chase controller 10 has a function for performing low-power long-distance wireless communication conforming to LoRa, and is capable of establishing a communication path with the edge device 30. The communication I / F 150B2 has a function for performing communication via a wireless LAN such as Wifi (registered trademark) (such as an access point function). The input / output I / F 140B is connected to the drone station 28. The memory unit 160B stores a chase control program for realizing the various functions described below.

[0033] The drone station 28 is capable of storing the drone 21 inside, and when the drone 21 takes off or lands, the takeoff and landing door opens and closes in response to a command from the chase controller 10 via the input / output I / F 140B (see Figure 1).

[0034] The drone 21 includes a drone body 24 and an associated LED 25b as a light emitting device and a directional speaker 26b as a sound emitting device. The drone body 24 is connected to a drone flight controller 23 connected to an on-board computer 22. The LED 25b is connected to an LED driver 25a, and the speaker 26b is connected to a sound emitting controller 26a. The drone flight controller 23, LED driver 25a, and sound emitting controller 26a may be separate from or integrated with the on-board computer 22. The communication I / F 150C of the on-board computer 22 has a function for communicating via a wireless LAN such as Wi-Fi, enabling communication with the repelling controller 10. A memory unit 160C stores a program for performing the repelling behavior described below.

[0035] (4) Functional Configuration of Edge Device 30 5 is a functional block diagram showing the functional configuration of the edge device 30 of the first embodiment. As shown in FIG. 5, the edge device 30 includes an imaging unit 32, a bird / animal stationing determination unit 33, a bird / animal position data acquisition unit 34, and an edge communication unit 35. Each of these units can be specifically realized by the processor 110A executing a program 68 related to the edge device stored in the storage unit 160A. Some or all of the functions of the edge communication unit 35 may be performed by the communication I / F 150A.

[0036] 6 is a sequence diagram illustrating the process from acquisition of image data 60 by the edge device 30 to step S22 in which the repelling controller 10 issues a command to repel birds and animals. Note that a data flow diagram is also used in part of the diagram (the same applies to the following sequence diagrams). The following continues to explain the functions and operations of each part of the edge device 30 with reference to FIGS. 5 and 6.

[0037] The imaging unit 32 captures the image of the monitoring area MA jAn image of the area including the image is captured to obtain image data 60 (S2). The image data 60 is temporarily stored in storage unit 160A. In the first embodiment, imaging unit 32 is configured with camera 32a.

[0038] The bird and animal stopping determination unit 33 determines the target bird and animal TBD, which is the target type of bird and animal, based on the image data 60 acquired by the imaging unit 32. j is the monitoring area MA j It is determined whether the bird is resting in a location (S4). Specifically, an inference model may be established in advance by deep learning for the target bird species (trained model 65 for identifying target bird and animal), image data 60 may be input to the trained model 65 for identifying target bird and animal, and bird recognition may be performed by evaluating the likelihood of the target bird species. In addition to this, another realization method, for example, if the external shape of the target bird and animal species is distinctive, pattern matching may be used to identify the target bird and animal TBD. j Recognition of the following may also be performed. If it is detected that the target bird or animal is staying in the same position for a predetermined period of time or more, the target bird or animal TBD j It may be determined that the object is stationary.

[0039] The bird and animal position data acquisition unit 34 acquires the target bird and animal TBD j The position data 61 of the target bird or animal TBD is acquired based on the image data 60 (S6). j However, the relative position of the target bird or animal in the image may be detected, and the position data 61 may be acquired based on the coordinates of the relative position. j The position data 61 is temporarily stored in the storage unit 160A.

[0040] The edge communication unit 35 is a target bird and animal TBD j The predetermined bird and animal stopping information 62 including the location data 61 is transmitted to the outside (S8).

[0041] 7 is a diagram showing an example of the data structure of the bird and animal stopping information 62. The bird and animal stopping information 62 includes at least the target bird and animal TBD jAs shown in the figure, the location data 61 of each target bird / animal is included, along with the identification ID of the edge device 30, the time when the target bird / animal was detected, the number of target birds / animals, etc. j The bird and animal stopping information 62 may be generated anew each time a group of target bird and animal TBDs or a single target bird and animal TBDs are detected.

[0042] (5) Functional configuration of the chase controller 10 Fig. 8 is a functional block diagram showing the functional configuration of the chase controller 10 of embodiment 1. As shown in Fig. 8, the chase controller 10 includes at least a repulse command unit 14, a repulse effect calculation unit 15, and an optimization processing unit 16. When the system includes an edge device 30, the chase controller 10 may further include a buffering unit 11 and a repulsion necessity determination unit 12. Each of these units can be specifically realized by having the processor 110B execute a chase control program 78 stored in the memory unit 160B.

[0043] (5-1) Determine whether there are any birds or animals that need to be repelled. (a) Although not shown in the figure, when detecting birds and animals using an area sensor that detects blockage using infrared rays, it may be determined that birds and animals that need to be repelled are stationary by detecting events where the blockage occurs intermittently at short intervals or continuously for a predetermined period of time or more. The area sensor can be connected to the input / output I / F 140B as appropriate.

[0044] (b) Alternatively, when bird and animal detection is performed using the imaging unit 32 of the edge device 30 as described above, the buffering unit 11 and the repelling necessity determination unit 12 of the repelling controller 10 may determine whether or not a bird or animal that needs to be repelled is present. The buffering unit 11 receives the bird and animal staying information 62 transmitted by the edge communication unit 35, and stores the bird and animal staying information 62 in the storage unit 160B in sequence (see also S12 in FIG. 6). The repelling necessity determining unit 12 determines whether or not there are birds and animals that need to be repelled staying within the monitoring area MA based on the bird and animal staying information 62 (see also S14 in FIG. 6).

[0045] (5-2) Order to repel When it is determined that a bird or animal that needs to be repelled is staying within the monitoring area, the repel command unit 14 causes the repelling device 20 to perform a repelling action based on the location information 71 of the staying bird or animal and the recommended repelling parameters 75 to be used during the sampling period (see also S14, S22, and S20 in FIG. 6). Specifically, the repelling action is performed while the location information 71 of the bird or animal and the recommended repelling parameters 75 are passed to the drone 21.

[0046] When the area sensor detects a bird or animal as in (a) above, the "position information 71 of the bird or animal that is staying" is the monitoring area MA j The ID (an appropriately assigned number) of the area sensor provided for each bird or animal may be used as the "location information of the bird or animal that is resting." In the case of (b) above, the "location information of the bird or animal that is resting" may be the location data 61 of the target bird or animal, or some coordinate data obtained from the image data 60, or data converted so that the repelling device 20 can understand it.

[0047] The "recommended avoidance parameter 75" is the sampling period SP j The recommended avoidance parameters 75 are stored in the storage unit 160B of the chase controller 10 and can be referenced as needed.

[0048] 9A and 9B are diagrams showing an example of data passed from the repelling controller 10 to the drone 21 when a repelling command is issued in embodiment 1. Fig. 9A shows an example of data on bird and animal position information 71, and Fig. 9B shows an example of data on recommended repelling parameters 75, which are exemplified by physical quantities that match the definitions of the repelling parameters shown in Fig. 2.

[0049] "Sampling period SP j" is the period for obtaining one sample of data when performing Gaussian regression, which will be described later. The length of the sampling period can be set as appropriate, but for example, one day may be used as the unit of the sampling period. Also, for example, one sampling period may be divided by the number of times the repelling behavior was performed.

[0050] (5-3) Implementation of repelling actions FIG. 10 is a sequence diagram that mainly explains the overall repelling action by the repelling device 20. When the repelling device 20 receives a repelling command from the repelling controller 10, it performs the repelling action in an appropriate manner. For example, this can be performed as shown in FIG. 10 (also see FIGS. 1, 4, etc.). That is, the repelling controller 10 inquires of the drone station 28 about operable drones (S82), and if an operable drone is present, issues a station open command (S83). In response, the drone station 28 opens its takeoff and landing doors (S84). When the drone 21 flies out of the drone station 28, the repelling controller 10 issues a repelling command to the drone 21 via the communication I / F 150B2, attaching bird / animal position information 71 and recommended repelling parameters 75 (S22). While the drone 21 performs an equipment inspection immediately before flight (S85), the onboard computer 22 on the drone 21 creates a specific flight plan and stimulus application plan based on the received bird / animal position information 71 and recommended repelling parameters 75, and departs for the repelling action in accordance with this plan (S86). The drone 21 (also equipped with an LED 25b, a speaker 26b, etc.) serving as the repelling device 20 stimulates the target bird or animal TBD within a planned range centered on the location where the bird or animal is stationed. When the drone 21 completes the repelling action and lands inside the drone station 28 (S87), the drone 21 or the drone station 28 reports the completion of the repelling action to the repelling controller 10 (S88), and the drone station 28 closes the takeoff and landing door (S89).

[0051] Fig. 11 is a sequence diagram mainly explaining the repelling effect calculation step S30 and the optimization processing step S40 by the repelling controller 10. Fig. 12 is a diagram for explaining an example of the bird and animal repelling system 1 and the bird and animal repelling method of the first embodiment. The following describes the repelling effect calculation unit 15 and the optimization processing unit 16 with reference to FIGS. 11 and 12 (also see FIG. 8).

[0052] (5-4) Calculation of the effect of repelling The chasing effect calculation unit 15 calculates the chasing effect during the sampling period SP i When the sampling period SP i The repelling effect of the repelling action carried out res The calculated repelling effect y is calculated (S30). res is temporarily stored in the storage unit 160B. res is a quantitative number, not a qualitative one.

[0053] The repelling effect calculation unit 15 may be configured as follows: First, a predetermined index value is calculated for each sampling period SP based on information about birds and animals obtained from image data of an area including the monitoring area MA. Examples of the "predetermined index value" that can be used include the number of target birds and animals TBD that have flown into the monitoring area MA, the number of "events" consisting of new arrivals and corresponding repelling actions, the average event interval, and the average time it takes for the target birds and animals to leave the monitoring area MA when repelling actions are carried out.

[0054] Next, the index value is calculated for each sampling period SP, and then the index value is calculated for the previous sampling period SP. i-1 The index value corresponding to the sampling period SP i By comparing the index value corresponding to res For example, if the sampling period SP is set in units of one day and the number of birds arriving in one day is used as the index value, then the previous sampling period SP i-1 Based on the number of birds arriving (the previous day) during the sampling period SP iThe fewer the number of birds arriving (today), the more effective the repelling effect. res Also, if the number of events per day is used as the index value, the greater the decrease in the number of events compared to the previous day, the greater the repelling effect y res It can also be defined as being large.

[0055] In the above case, the chase-away effect calculation unit 15 calculates the chase-away effect y based on the bird and animal stationary information 62 buffered in the storage unit 160B of the chase-away controller 10. res It is preferable to calculate

[0056] In addition, rather than the relative comparison on a sampling period basis as described above, the repelling effect is calculated by a relative comparison on an event basis. res For example, the repelling effect y can be calculated by quantifying the percentage of birds repelled within a given time period for one event. res In any case, it is possible to calculate the repelling effect y based on information about birds and animals obtained from image data capturing an area including the monitoring area MA. res It is possible to calculate

[0057] (5-5) Optimization process (a) Addition of new implementation data 72 to the regression data storage unit 18 The optimization processing unit 16 first calculates the recommended repelling parameter 75(X) used in the repelling behavior during the sampling period and the repelling effect y calculated by the repelling effect calculation unit 15. res The resulting working data 72 is added to the regression data storage unit 18 as new regression data (S41). The working data 72 is an individual set of input / output data used in Gaussian process regression of Bayesian optimization, which will be described later, and can also be considered as data of each sample point from the perspective of Gaussian process regression.

[0058] Fig. 13 is a diagram for explaining data used in the optimization processing unit 16. Fig. 13(a) shows an example of the implementation data 72(x, y), and Fig. 13(b) shows an example of the data structure of the regression data accumulation unit 18. As shown in FIG. 13(a), the field of the avoidance parameter X of the implementation data 72 contains the sampling period SP i The recommended repellent parameters 75 used in the repelling action of the sampling period SP are copied. i The repelling effect of the repelling action carried out res is copied. The regression data accumulation unit 18 is stored in the memory unit 160B, and is constructed by adding one set of input / output data 72 (X, Y) to the execution data 72 at the end of each sampling period SP (see Figure 13(b)).

[0059] (b) Bayesian optimization FIG. 14 shows the optimization process performed by the optimization processor 16 of the first embodiment to calculate the avoidance parameter x corresponding to the maximum effect point MEP. next is a schematic distribution map used to explain how the repelling effect y is derived. In reality, there are many repelling parameters X (x1, x2, x3, ...) that can affect the repelling effect y, and they are thought to form a multidimensional space in relation to the repelling effect Y. However, for ease of understanding, we will assume in Figure 14 that only two parameters x1 and x2 can affect the bird repelling effect y, and continue the explanation using a distribution map (response function) that represents the space of two-dimensional inputs (repelling parameters x1, x2) and the corresponding output (repelling effect y).

[0060] The Bayesian optimization method is a method for efficiently finding the maximum value of an unknown function. As described above, there are many parameters as candidates for the repellent stimulus to be presented to birds, which constitute a multidimensional space. However, as can be seen in Figure 14, there is at least one combination of repellent parameters X that birds dislike the most. In this specification, the point that birds dislike the most, that is, the point that is thought to have the greatest repellent effect (in other words, the point where the acquisition function related to Bayesian optimization is maximized), is called the "maximum effectiveness point MEP."

[0061] There must exist an optimal combination of repellent stimuli (repellent parameters) for repelling birds. However, as mentioned above, due to the multidimensional nature of the function, searching for such parameters is generally difficult. If a method were adopted in which repellent parameters are sequentially changed while comprehensively presenting repellent stimuli, as in classical experimental design, bird damage would not be prevented while stimuli with low repellent effectiveness are presented. Therefore, it is necessary to find the optimal stimulus as quickly as possible, taking into account the birds' habituation to the stimuli. Bayesian optimization creates a regression model of the target function from past observations using Gaussian process regression, etc., obtains an acquisition function based on this regression model, and can identify the point at which the acquisition function's value is maximized (the maximum effectiveness point, or MEP). This allows the optimal combination of input parameters (optimal solution; in this case, the repellent parameter X) to be derived.

[0062] The Bayesian optimization process can be broadly divided into Gaussian process regression, updating the posterior probability distribution and acquisition function, searching for the maximum effect point of the acquisition function, and deriving the optimal solution corresponding to the maximum effect point. Bayesian optimization uses a sequential recency model based on Gaussian process regression to search for the maximum effect point, allowing for high-speed searches. In addition, regression can be performed even with a relatively smaller number of samples than other algorithms, making it possible to efficiently find the optimal solution.

[0063] (c) Bayesian optimization process The optimization processing unit 16 executes Bayesian optimization processing based on the group of implementation data 72 stored in the regression data storage unit 18, and thereby obtains the avoidance parameter x corresponding to the point (maximum effect point MEP) at which the acquisition function related to the Bayesian optimization is maximized. next Specifically, Gaussian process regression is performed based on 72 groups of implementation data stored in regression data storage unit 18 (S41), the maximum effect point MEP is searched for (S44), and the value of the avoidance parameter corresponding to the identified maximum effect point MEP is derived (S46). Then, the optimization processing unit 16 calculates the derived new avoidance parameter x next is set as the recommended avoidance parameter (see Figure 11).

[0064] The optimization processing unit 16 includes at least the elements of a Gaussian process regression executing unit 16a, a maximum effect point searching unit 16b, a next avoidance parameter deriving unit 16c, and a recommended avoidance parameter updating unit 16d (see FIG. 8).

[0065] The Gaussian process regression execution unit 16a performs Gaussian process regression based on the group of experimental data 72 stored in the regression data storage unit 18 (S41). At this time, it is preferable that the optimization processing unit executes Bayesian optimization processing based on a predetermined set of experimental data 72, including new experimental data 72 corresponding to the most recent sampling period SPi. In other words, it is preferable to perform Gaussian process regression using a sample data group that always includes the most recently generated new experimental data 72. For example, in FIG. 13(b), Gaussian process regression is performed using a sample data group including experimental data 72 corresponding to data set No. 9, indicated as "latest experimental data 72." In the first embodiment, Gaussian process regression is performed using all experimental data 72 stored in the regression data storage unit 18, indicated by D1.

[0066] When new implementation data 72 is added and the Gaussian process regression is completed, the distribution of the response function that represents the relationship between the input X (avoidance parameter) and the output Y (repelling effect) is also updated. i-1 "Day Distribution (Shape)"i The "future distribution (shape)" changes. Accordingly, the distribution (shape) of the acquisition function based on the response function is also updated. The acquisition function is a function that expresses the degree of possibility of improving beyond the previous optimal value. The acquisition function can be constructed in an appropriate manner. For example, the expected value and standard deviation of the response function can be taken into account to create the acquisition function.

[0067] The maximum effect point search unit 16b searches for a point at which the acquisition function related to the Bayesian optimization is maximized. For example, in the image of Figure 14, the maximum effect point search unit 16b searches for and identifies the maximum effect point indicated by the symbol MEP. Such a search for the maximum effect point MEP can also be realized by utilizing library software incorporated in the chase control program stored in the memory unit 160B of the chase controller 10, for example.

[0068] The next avoidance parameter derivation unit 16c derives the value of the avoidance parameter corresponding to the identified maximum effect point MEP. For example, in the image of FIG. 14, the next avoidance parameter derivation unit 16c derives specific numerical values ​​x1, x2 on each axis of the avoidance parameters x1, x2 of the input system corresponding to the identified maximum effect point MEP. next ,x2 next is derived by back calculation.

[0069] (d) Update of recommended avoidance parameter 75 As shown in FIG. 12, the recommended avoidance parameter update unit 16d updates the avoidance parameter X derived by the next avoidance parameter derivation unit 16c. next The value is set (assigned) to the recommended avoidance parameter 75 (S48).

[0070] (5-6) Next sampling period SP i+1 Repelling action in Next sampling period SP i+1 In the example, the recommended avoidance parameter 75 (X i+1 Based on this, the repelling command unit 14 causes the repelling device 20 to perform repelling action.

[0071] 2. Effects of the bird and animal repelling system 1 and repelling method according to the first embodiment (1) The bird and animal repelling system 1 according to the first embodiment i The recommended repellent parameters used in i ) and the sampling period SP i The repelling effect of res After adding the implementation data (x, y) consisting of the above to the regression data storage unit 18, Bayesian optimization is performed based on the implementation data 72 groups stored in the regression data storage unit 18, and the avoidance parameter x corresponding to the point (maximum effect point MEP) at which the acquisition function related to Bayesian optimization is maximized is calculated. next and derive the new repellent parameter x next is set as the recommended avoidance parameter 75, and the next sampling period SP i+1 In the new recommended parameter 75 (X i+1 ) the repelling command unit 14 is configured to cause the repelling device 20 to perform repelling action (see FIG. 12).

[0072] In this way, evaluation is performed every time the sampling period SP ends, and the Bayesian optimization is also performed by regression including the most recent implementation data 72. Therefore, the maximum effect point MEP searched by the optimization processing unit 16 and the corresponding avoidance parameter x next The most recent changes in conditions (such as birds becoming accustomed to a certain pattern of repellent stimulus) are also reflected in the next sampling period SP. i+1 In this case, the latest repellent parameter (X next ) to repel the enemy.

[0073] Therefore, the repellent effect can be maintained without the birds becoming bored with the repellent stimulus. In addition, even if conditions change, such as when a previous flock of birds is replaced by a new flock or when a nearby structure is moved, the repellent parameter (X next) can be quickly proposed, and a decline in the repelling effect can be prevented. Furthermore, since the optimization processing unit 16 proposes the repelling parameters that are considered to be most effective, the repelling effect y can be maintained at a high level.

[0074] As described above, the bird and animal repellent system 1 according to the first embodiment has a highly durable repellent effect.

[0075] Furthermore, the bird and animal repellent system 1 according to the first embodiment utilizes a Bayesian optimization technique to determine the optimal repellent parameter x next Bayesian optimization is said to quickly converge the regression curve (here, the response function of the repelling effect y to the repellent parameter x) even with a relatively small number of samples. This makes it possible to quickly find a highly effective combination pattern of repellent methods (repellent parameters) with a small number of trials, without the inconvenience of conventional machine learning where the control period ends before an estimation model adapted to the birds that have arrived is established and bird repelling measures cannot be implemented in time.

[0076] (2) The optimization processing unit 16 calculates the “most recent” sampling period SP i The Bayesian optimization process is performed based on a predetermined set of experimental data 72 including new experimental data 72 corresponding to the above (see FIGS. 12 and 13(b)). In other words, when performing Gaussian process regression for Bayesian optimization, the most recent sampling period SP i Since the system also reflects information about changes in conditions in the environment, it can be adapted with minimal delay to, for example, changes in the bird's degree of habituation to an aversive stimulus.

[0077] (3) The repelling effect calculation unit 15 calculates a predetermined index value based on information about birds and animals obtained from image data 60 capturing an area including the monitoring area MA, and calculates the repelling effect of the previous sampling period SP i-1 and the corresponding index value for the sampling period SP iThe repelling effect y may be calculated by comparing an index value corresponding to with . In this way, the repelling effect y can be calculated quantitatively, rather than grasping a discrete repelling effect such as success / failure. Furthermore, since the index value is based on information about birds and animals obtained from the image data 60, the image data 60 from the imaging unit 32 can be effectively used, and it is also easy to automate the calculation of the effect.

[0078] (4) The bird and animal repelling system 1 further includes an edge device 30, which captures image data 60 using an imaging unit 32 of the edge device 30, determines whether the bird or animal is stationary, acquires position data of the target bird or animal, and transmits bird and animal stationing information 62. In response to this, the repelling controller 10 buffers the bird and animal stationing information 62, determines whether any bird or animal that needs to be repelled is stationary based on the bird and animal stationing information 62, and requests the repelling command unit 14 to dispatch a unit to repel the bird or animal. With this configuration, it is possible to identify the bird and animal species using the edge device 30, and to repel them using repellent parameters for each bird and animal species. It is also possible to identify the detailed coordinates (position data 61) of the bird and animal from the image data 60, and to repel them accurately based on those coordinates. It is also possible to automatically calculate the repelling effect y based on image data 60 from the imaging unit 32, which not only reduces the effort required compared to manually calculating the repelling effect, but also shortens the time lag in calculation time, contributing to the continuation of the repellent effect.

[0079] 3. Bird and animal repellent method and program according to embodiment 1 Next, a method and program for repelling birds and animals according to the first embodiment will be described with reference to FIGS. 6, 10, 11 and 12. FIG.

[0080] (1) The bird and animal repelling method of embodiment 1 involves updating an "avoidance parameter X" that quantitatively indicates the type of stimulus that repels birds and animals for each sampling period SP, and repelling birds and animals in accordance with the avoidance parameter X, and includes the following steps: When it is determined that a bird or animal that needs to be repelled is staying within the monitoring area MA, the location information 71 of the staying bird or animal and the sampling period SP i a repelling command step S22 that causes the repelling device 20 to perform repelling action based on the recommended repelling parameters 75 to be used; The sampling period SP i When the sampling period SP i A repelling effect calculation step S30 calculates the repelling effect y of the repelling action carried out. The sampling period SP i Recommended repellent parameter X used in repelling action i The implementation data (x, y) consisting of the repulsion effect y calculated in the repulsion effect calculation step S30 is added to the regression data storage unit 18 as new regression data (S41), and then Bayesian optimization processing is performed based on the implementation data group stored in the regression data storage unit 18, and the repulsion parameter x corresponding to the point (maximum effect point MEP) at which the acquisition function related to the Bayesian optimization is maximized is calculated. next (S42, S44, S46), and the derived new avoidance parameter x next is set as the recommended avoidance parameter 75 (S48) in the optimization processing step S40. Then, the next sampling period SP i+1 In the repel command step S22, the newly set recommended repel parameters (x next ) and causes the repelling device 20 to perform repelling action.

[0081] (2) The repelling command unit 14, the repelling effect calculation unit 15, the optimization processing unit 16, etc., which constitute the repelling controller 10 of the bird and animal repelling system 1 according to the first embodiment, and the series of processes such as the repelling command step S22, the repelling effect calculation step S30, and the optimization processing step S40, can be executed by hardware incorporating those functions, or by software. When the series of processes are executed by software, this can be realized by having the processor 110B constituting the repelling controller 10 execute a program (repelling control program; see FIG. 4) stored in the memory unit 160B.

[0082] The program of embodiment 1 is a program used in a bird and animal repelling system 1 that includes at least a repelling controller 10 that controls the repelling of birds and animals and a repelling device 20 that performs repelling actions on birds and animals to be repelled in response to commands from the repelling controller 10, and that updates an "avoidance parameter X" that quantitatively indicates the type of stimulus that repels birds and animals for each sampling period SP, and repels birds and animals in accordance with the repelling parameter X. This program causes the computer (processor 110B) constituting the repelling controller 10 to perform the repelling command step S22, the repelling effect calculation step S30, and the optimization processing step S40. In the repelling command step S22 in the next sampling period, the newly set recommended repelling parameter (x next ) and causes the repelling device 20 to perform repelling action and processing. The details of the repelling command step S22, the repelling effect calculation step S30, and the optimization processing step S40 can be incorporated as components of the program by referring to the explanations in 2(1) above.

[0083] Since the main parts of the bird and animal repelling method and program of embodiment 1 are common to the main parts of the bird and animal repelling system 1 of embodiment 1, the effects of the bird and animal repelling method and program can be similar to the effects of the bird and animal repelling system 1 of embodiment 1.

[0084] [Embodiment 2] The bird and animal repelling system 2 (not shown) of embodiment 2 basically has the same configuration as the bird and animal repelling system 1 of embodiment 1, but differs from the bird and animal repelling system 1 of embodiment 1 in the way in which the regression data is referenced in the optimization processing unit 16.

[0085] That is, in the bird and animal repelling system 2 (not shown) according to the second embodiment, the optimization processing unit 16 calculates the sampling period SP i-j From the most recent sampling period SP i The Bayesian optimization process is performed based only on the implementation data 72 corresponding to the period up to the present time.

[0086] 13, in the first embodiment, Bayesian optimization is performed using all of the experimental data 72 (the group of experimental data 72 in the range indicated by D1) stored in the regression data storage unit 18. On the other hand, in the second embodiment, Bayesian optimization is performed using, for example, the group of experimental data 72 in the range indicated by D2.

[0087] For example, in Figure 13, if the sampling period is set in days, the sampling period SP 5 days ago iー5 from the most recent sampling period SP i Bayesian optimization Gaussian process regression is performed using only the implementation data corresponding to the period up to (today). In other words, the implementation data 72 corresponding to the sampling period older than the specified period (the sampling period older than 6 days ago) iー6 Previously corresponding implementation data72) will be actively discarded and not used in Gaussian process regression.

[0088] This allows the optimum repelling parameter x to be calculated based on the implementation data 72 that reflects the current state of the bird flock, without being influenced by the implementation data 72 that reflects the past state of the bird flock. next This makes it possible to derive a model that continues to maintain its repellent effect even when the flock of birds becomes accustomed to the bird or when the birds are replaced by a new flock, because the model is based on the most recent data.

[0089] The bird and animal repelling system 2 according to the second embodiment has basically the same configuration as the bird and animal repelling system 1 according to the first embodiment, except for the way in which regression data is referenced in the optimization processing unit 16. Therefore, the bird and animal repelling system 2 has the same effects as the bird and animal repelling system 1.

[0090] [Embodiment 3] The bird and animal repelling system 3 (not shown) of embodiment 3 basically has the same configuration as the bird and animal repelling systems 1 and 2 of embodiments 1 and 2, but differs from the bird and animal repelling systems 1 and 2 of embodiments 1 and 2 in the way in which the optimization processing unit 16 extracts search points.

[0091] That is, in the bird and animal repellent system 3 (not shown) according to the third embodiment, under predetermined conditions, the optimization processing unit 16 extracts a random point other than the maximum effect point MEP, and calculates the repellent parameter x corresponding to the random point. next and deriving the derived new repellent parameter xnext is configured to set as a recommended avoidance parameter. When the posterior probability distribution obtained by the acquisition function or Bayesian optimization is regarded as a distribution map, a random point may be extracted from another peak Mt2, Mt2, etc., which has a peak other than the peak Mt1 containing the peak considered to be the maximum effect point MEP, as shown in FIG. 14, for example.

[0092] Here, "predetermined conditions" may be once every several samplings (for example, if sampling is performed for 10 days, sampling based on embodiment 3 is performed on one day), or periodically / irregularly, or when it is determined that the repelling effect y is on a medium-term downward trend.

[0093] In this way, rather than just pursuing the maximum effective point MEP at that time, by occasionally (randomly) hitting points outside the maximum effective point MEP and carrying out repelling actions with the repelling parameters corresponding to the random points, if the system has fallen into a local solution, it is possible to escape from the local solution and find the optimal solution when viewed over a wider area. Also, by selecting a random point from another mountain with a different peak, such as Mt2 or Mt3, it may be possible to attempt repelling on a mountain whose repelling effect y has recently been increasing. In such cases, by trying to do so before the mountain rises, it is possible to seize an opportunity for timely repelling.

[0094] Although the present invention has been described above based on the above embodiment, the present invention is not limited to the above embodiment and can be embodied in various forms without departing from the spirit of the present invention, and for example, the following modifications are also possible.

[0095] (1) In each embodiment, an example has been described in which one set of repelling devices 20 is used, but the present invention is not limited to this. Multiple sets of repelling devices 20 (drones 21, etc.) may be provided. By providing multiple sets of repelling devices 20, even if birds and animals invade multiple monitoring areas MA at the same time, it is possible to divide the work and carry out repelling actions in each monitoring area in parallel.

[0096] (2) In FIG. 4, the first embodiment is described with a configuration in which one edge device 30 is shown, but the present invention is not limited to this. As shown in FIG. 1, a plurality of edge devices 30 are provided, and each edge device 30 -1 ~30 -5It is also possible to adopt a configuration in which the above-mentioned devices are connected to a single common chase controller 10 via wireless communication means.

[0097] (3) In a system having a plurality of image capturing units 32, the repelling controller 10 is provided with a group of implementation data 72 as regression data and a recommended repelling parameter 75 (data container) corresponding to each image capturing unit 32, the repelling effect calculation unit 15 calculates the repelling effect y for each image capturing unit 32, and the optimization processing unit 16 executes Bayesian optimization processing for each image capturing unit 32, and the obtained new repelling parameter x next may be configured to be set as the recommended avoidance parameters 75 corresponding to each imaging unit 32.

[0098] In this way, by accumulating a group of implementation data 72 for each monitoring area MA corresponding to each imaging unit 32 and performing optimization processing for each, the avoidance parameter x suitable for each monitoring area MA can be obtained. next This will make it possible to carry out detailed repelling actions.

[0099] (4) In each embodiment, the chase controller 10 is implemented by one information processing device 100, but the present invention is not limited to this. A large-scale field may be divided into areas, and an information processing device 100 may be provided for each area. Furthermore, multiple chase controllers 10 may be provided.

[0100] (5) In each embodiment, the optimization processing unit 16 calculates the next sampling period SP i+1 The avoidance parameter x to be implemented next In the above example, the derivation of is performed using a Bayesian optimization technique. However, the present invention is not limited to this. The optimization processing unit 16 can also be configured using an optimization technique similar to Bayesian optimization or other optimization techniques, and such configurations are also considered equivalent to the present invention.

[0101] (6) In each embodiment, birds flying into grape farms have been described as an example, but the present invention is not limited to this. The present invention can be applied to places other than grape farms, and other animals (deer, boars, monkeys, bears, raccoons, palm civets, etc.) can also be applied as target birds and animals TBD. Furthermore, the present invention is not limited to birds and animals as long as the target can achieve the effects of the present invention. [Explanation of symbols]

[0102] 1, 2, 3...Bird and animal repelling system, 10...repel controller, 11...buffering unit, 12...repel necessity determination unit, 14...repel command unit, 15...repel effect calculation unit, 16...optimization processing unit, 16a...Gaussian process regression execution unit, 16b...maximum effect point search unit, 16c...next repellent parameter derivation unit, 16d...recommended repellent parameter update unit, 18...regression data storage unit, 20...repel device, 21...drone, 22...on-board computer, 23...drone flight controller, 24...drone main body, 25a...LED driver, 25b...LED, 26a...sound emission controller, 26b...speaker, 28...drone station, 30...edge device, 30...edge device, 32...imaging unit, 32a...camera, 33...bird and animal station determination unit, 34...bird and animal position Position data acquisition unit, 35...edge communication unit, 60...image data, 61...position data of target birds and animals, 62...bird and animal stopping information, 65...trained model for identifying target birds and animals, 68...program for edge device, 71...bird and animal position information, 72...implementation data, 75...recommended repellent parameters, 78...chasing control program, 100, 100A, 100B, 100C...information processing device, 1 10, 110A, 110B, 110C...processor, 120...memory, 130...storage, 140, 140A, 140B...input / output I / F, 150, 150A, 150B1, 150B2, 150C...communication I / F, 160, 160A, 160B, 160C...memory unit, BD...birds and animals, TBD...target birds and animals, MA...monitoring area, MEP...maximum effect point, Mt1, Mt2, Mt3...mountains

Claims

1. A bird and animal repelling system that updates an "avoidance parameter" that quantitatively indicates the type of stimulus that repels birds and animals for each sampling period, and repels the birds and animals in accordance with the avoidance parameter, The system includes at least a repelling controller that controls the repelling of the birds and animals, and a repelling device that receives a command from the repelling controller and performs a repelling action against the birds and animals to be repelled, The chase controller a repelling command unit that, when it is determined that the bird or animal that needs to be repelled is staying within the monitoring area, causes the repelling device to carry out the repelling action based on the location information of the bird or animal that is staying and the recommended repelling parameters to be used during the sampling period; a repulsion effect calculation unit that calculates a repulsion effect y of the repulsion behavior performed during the sampling period when the sampling period ends; The implementation data (x, y) consisting of the recommended repellent parameter x used in the repelling behavior during the sampling period and the repelling effect y calculated by the repelling effect calculation unit is added to a regression data storage unit as new regression data, and then a Bayesian optimization process is performed based on the implementation data group stored in the regression data storage unit, and the repelling parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized is calculated. ｎｅｘｔ and derive the new avoidance parameter x ｎｅｘｔ an optimization processing unit that sets the recommended avoidance parameter as Equipped with In the next sampling period, the repelling controller causes the repelling command unit to cause the repelling device to perform the repelling action based on the recommended repelling parameters newly set by the optimization processing unit. A bird and animal repelling system.

2. The bird and animal repellent system according to claim 1, The optimization processing unit performs the Bayesian optimization process based on a predetermined set of implementation data including new implementation data corresponding to the most recent sampling period.

3. 3. The bird and animal repelling system according to claim 2, The optimization processing unit performs the Bayesian optimization process based only on the implementation data corresponding to the period from the sampling period going back a predetermined number of periods to the most recent sampling period.

4. The bird and animal repelling system according to any one of claims 1 to 3, The repulsion effect calculation unit calculating a predetermined index value based on information about the bird or animal obtained from image data of an area including the monitoring area; The repelling effect y is calculated by comparing the index value corresponding to the previous sampling period with the index value corresponding to the current sampling period. A bird and animal repelling system.

5. 5. The bird and animal repellent system according to claim 4, further comprising an edge device; The edge device an imaging unit that captures an image of an area including the monitoring area and acquires the image data; a bird and animal resting determination unit that determines whether a target bird or animal, which is the bird or animal of the target type, is resting in the monitoring area based on the image data acquired by the imaging unit; a bird / animal position data acquisition unit that acquires position data of the target bird / animal that is stationary based on the image data; an edge communication unit that transmits predetermined bird and animal stop information including the position data to an external device; The chase controller a buffering unit that receives the bird and animal staying information transmitted by the edge communication unit and sequentially stores the bird and animal staying information in a storage unit; and a repelling necessity determining unit that determines whether or not the bird or animal that needs to be repelled is staying within the monitoring area based on the bird or animal staying information, and requests the repelling command unit to be dispatched to repel the bird or animal if it determines that the bird or animal that needs to be repelled is staying. A bird and animal repelling system.

6. 6. The bird and animal repellent system according to claim 5, The bird and animal repelling system is characterized in that the repelling effect calculation unit calculates the repelling effect y based on the bird and animal stationary information buffered in the memory unit.

7. 7. The bird and animal repellent system according to claim 5 or 6, The imaging unit includes a plurality of imaging units, the repelling controller is provided with the implementation data group and the recommended avoidance parameters as regression data corresponding to each of the imaging units, the repulsion effect calculation unit calculates a repulsion effect y for each of the imaging units, The optimization processing unit executes the Bayesian optimization process for each imaging unit, and sets the obtained new repellent parameters as the recommended repellent parameters corresponding to each imaging unit.

8. The bird and animal repelling system according to any one of claims 1 to 7, If the point where the value of the acquisition function is the maximum is the maximum effect point, Under a predetermined condition, the optimization processing unit extracts a random point other than the maximum effect point, and calculates the avoidance parameter x corresponding to the random point. ｎｅｘｔ and derive the new repellent parameter x ｎｅｘｔ as the recommended avoidance parameter; A bird and animal repelling system.

9. 9. The bird and animal repellent system according to claim 8, A bird and animal repellent system characterized in that, when the acquisition function or the posterior probability distribution obtained by Bayesian optimization is viewed as a distribution map, the random point is extracted from another mountain having a peak separate from the mountain containing the peak considered to be the point of maximum effect.

10. A bird and animal repelling method, which updates an "avoidance parameter" that quantitatively indicates the type of stimulus that repels birds and animals for each sampling period, and repels the birds and animals in accordance with the avoidance parameter, a repelling command step for causing the repelling device to perform repelling actions based on the location information of the bird or animal that is staying within the monitoring area and the recommended repelling parameters to be used during the sampling period, when it is determined that the bird or animal that needs to be repelled is staying within the monitoring area; a repulsion effect calculation step of calculating a repulsion effect y of the repulsion behavior performed during the sampling period when the sampling period ends; The implementation data (x, y) consisting of the recommended repellent parameter x used in the repelling behavior during the sampling period and the repelling effect y calculated in the repelling effect calculation step is added to a regression data storage unit as new regression data, and then a Bayesian optimization process is performed based on the implementation data group stored in the regression data storage unit, and the repelling parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized is calculated. ｎｅｘｔ and derive the new repellent parameter x ｎｅｘｔ an optimization processing step of setting the recommended avoidance parameter as In the repelling command step in the next sampling period, the repelling device is caused to perform the repelling behavior based on the recommended repelling parameters newly set by performing the optimization processing step. A method for repelling birds and animals.

11. A program used in a bird and animal repelling system that includes at least a repelling controller that controls repelling of birds and animals and a repelling device that executes repelling behavior of the birds and animals in response to a command from the repelling controller, and that updates "repelling parameters" that quantitatively indicate the type of stimulus that repels birds and animals at each sampling period and repels the birds and animals in accordance with the repelling parameters, A computer constituting the chase controller, a repelling command step for causing the repelling device to perform repelling actions based on the location information of the bird or animal that is staying within the monitoring area and the recommended repelling parameters to be used during the sampling period, when it is determined that the bird or animal that needs to be repelled is staying within the monitoring area; a repulsion effect calculation step of calculating a repulsion effect y of the repulsion behavior performed during the sampling period when the sampling period ends; The implementation data (x, y) consisting of the recommended repellent parameter x used in the repelling behavior during the sampling period and the repelling effect y calculated in the repelling effect calculation step is added to a regression data storage unit as new regression data, and then a Bayesian optimization process is performed based on the implementation data group stored in the regression data storage unit, and the repelling parameter x corresponding to the point at which the acquisition function related to the Bayesian optimization is maximized is calculated. ｎｅｘｔ and derive the new repellent parameter x ｎｅｘｔ an optimization processing step of setting the recommended avoidance parameter as In the repelling command step in the next sampling period, the repelling device is caused to perform repelling behavior based on the recommended repelling parameters newly set by performing the optimization processing step. The program that performs the processing.

Images