Method and device for generating simulated echo image, method and device for generating estimator, method and device for estimating fish population, and program

Abstract

To allow the number of fish present in an underwater space to be estimated accurately.SOLUTION: In a simulation of a predetermined number of fish moving in a predetermined underwater space according to an equation of motion, a method for generating a simulated echo image simulating reflected waves when sound waves are transmitted into the underwater space includes calculating a required number of fish, which is the number of fish that need to be present in the simulated echo image based on an estimation accuracy of the predetermined number of fish, determining a cutout window size for the simulated echo image based on the required number of fish, and generating a simulated echo image with the determined cutout window size.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a simulated echo image generation method and apparatus thereof, an estimator generation method and apparatus thereof, a method and apparatus for estimating the number of fish individuals, and a program.

Background Art

[0002] Conventionally, there is an estimator for estimating the number of fish present in an underwater space by machine learning using, as teacher data, a plurality of learning data sets each including a learning echo image based on sound waves transmitted into the underwater space where fish are present and reflected and received by the fish, and the number of fish present in the underwater space in the echo image.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present disclosure is to provide a simulated echo image generation method and apparatus thereof, an estimator generation method and apparatus thereof, a method and apparatus for estimating the number of fish individuals, and a program that can accurately estimate the number of fish present in an underwater space.

Means for Solving the Problems

[0005] One aspect of the present disclosure is a method for generating a simulated echo image that simulates a reflected wave when a sound wave is transmitted into a space in water, in a simulation where a predetermined number of fish move in accordance with the equations of motion in the space in water, the method including: determining the size of a cutout window of the simulated echo image based on the required number of fish that need to exist in the simulated echo image, the required number being calculated based on the estimation accuracy of the predetermined number of fish; and generating the simulated echo image with the determined size of the cutout window.

[0006] The present disclosure includes a simulated echo image generation apparatus and a program having the same features as the above-described simulated echo image generation method.

[0007] One aspect of the present disclosure is a method for generating an estimator for a predetermined number of fish, using a simulation in which a predetermined number of fish move in accordance with the equations of motion in a space in water, the method including: obtaining a simulated echo image generated with a size of a cutout window determined based on the required number of fish that need to exist in the simulated echo image that simulates a reflected wave when a sound wave is transmitted into the space in water, the required number being calculated based on the estimation accuracy of the predetermined number of fish; and generating the estimator by learning using the simulated echo image.

[0008] The present disclosure includes an estimator generation apparatus and a program having the same features as the above-described estimator generation method.

[0009] One aspect of the present disclosure is to obtain an echo image showing a reflected wave when a sound wave is transmitted into a predetermined space in water, and to estimate the number of fish existing in the space in water by using the echo image and an estimator generated by learning a simulated echo image. Including, in a simulation where a predetermined number of fish move according to the equation of motion in a space in the predetermined water, the necessary number of fish that need to be present in a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the water space. The size of the cutout window is determined based on the necessary number calculated based on the estimation accuracy of the predetermined number of individuals, and the echo image is generated with the size of the cutout window. This is a method for estimating the number of fish individuals in which the echo image is generated with the size of the cutout window.

[0010] This disclosure includes an estimation device and a program having the same features as the above-described method for estimating the number of fish individuals.

Effect of the Invention

[0011] According to this disclosure, it is possible to accurately estimate the number of fish present in a space in water.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

[0013] The training echo images can be generated by transmitting sound waves into an underwater space in a simulation and generating echo images based on the sound waves reflected and received by an unknown number of fish present in the underwater space. By applying an estimator obtained by machine learning using the training echo images to images from a fish finder installed in the fish pen, the number of fish (number of tails, i.e., number of individuals) to be counted (referred to as target fish) present in the fish pen can be estimated.

[0014] The fishfinder image is, for example, an image obtained by searching a cone-shaped area directly below a transducer, and the school of fish observed in the image represents only a portion of the fish in the fish pen. In other words, to estimate the number of target fish, a cutout window is set in the fishfinder image, and the total number of target fish in the entire fish pen is estimated from an image cut out from the fishfinder image according to the cutout window (referred to as a cutout image).

[0015] The inventors of this application found that there is a relationship between the crop size of the image and the estimation accuracy. In other words, a certain number of target fish must be captured within the crop window. For example, when estimating the number of target fish in a fish tank containing 800 target fish, an error of about 10 fish (estimated To estimate (also called fixed error), consider how much estimation accuracy is required. The estimation accuracy in this case is 10÷800×100 = 1.25%. Also, since the counting unit is 1 fish, if the minimum number of target fish that needs to be reflected in the cutout window is X, then X×1.25%≧1, so X = 80. That is, it is necessary for there to be 80 or more target fish in the cutout image for counting with the required estimation accuracy. In other words, to estimate or predict about 800 target fish with an error unit of about 10, a cutout image in which 80 or more target fish are reflected is acquired, and the acquired cutout image is used for machine learning.

[0016] The size of the cutout window is determined by the depth range width (depth width) of the fish finder and the transmission period and number of sound waves (referred to as the beam rate and number of pings. That is, horizontal time). Also, the number of target fish reflected in the cutout window is related to the beam width of the transmitted and received sound waves. That is, the wider the beam width, the more target fish are detected in one transmission. Also, when estimating the number of target fish with an estimator (AI) obtained by machine learning, the conditions (size) of the cutout window related to the cutout image used as machine learning data are made the same as the conditions of the cutout window for the actual image obtained by the fish finder.

[0017] <Method for generating simulated echo image, simulated echo image generation device, and program> The simulated echo image generation method according to the embodiment is a method for generating a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space in a simulation where a predetermined number of fish move according to the equation of motion in a predetermined underwater space. The simulated echo image generation method includes determining the size of the cutout window of the simulated echo image based on the required number of individuals, which is the number of fish that need to exist in the simulated echo image and is calculated based on the estimation accuracy of the predetermined number of individuals, and generating the simulated echo image with the determined size of the cutout window. The simulated echo image generation method may be performed manually or the information processing device may execute each step. That is, the information processing device may operate as a simulated echo image generation device that implements the simulated echo image generation method. Also, a simulated echo image may be generated by executing a program on a computer.

[0018] The predetermined underwater space is, for example, the space inside a net cage or the like where the space below the water surface of open water (such as a river, lake, pond, sea, etc.) or underwater is partitioned using a net or the bottom. The underwater is formed by fresh water, brackish water, or seawater in which the fish to be counted (target fish) can inhabit. The underwater space may be not only using the natural environment but also the space inside an artificially water-stored structure such as a pool or an aquarium. In the underwater space, there are a predetermined number (predetermined number) of target fish. The target fish may be freshwater fish or seawater fish. The type of fish is, for example, a fish targeted by aquaculture or cultivation fishery. The target fish is, for example, a migratory fish such as tuna, amberjack, kampachi, mackerel, or saury. Also, the target fish may be non-migratory fish such as sea bream, pufferfish, flounder, or salmonids. The target fish is not limited to the above examples. The target fish may include aquatic organisms (fish and shellfish) such as shrimp in addition to fish. In this embodiment, an example where the target fish is tuna will be described.

[0019] In a live fish trap with a net set in water, the boundary between the inside and the outside is partitioned by the net, the water surface, the bottom, etc., so that fish cannot move back and forth between the inside and the outside of the fish trap. The shape of the space in the water can adopt various shapes such as a rectangular parallelepiped, a cube, a cylindrical shape, etc. As an example, it is conceivable to define the space in the water by a rectangular parallelepiped with length, width, and height. In this case, the space in the water is a rectangular parallelepiped space with a horizontal length X, a vertical length Y, and a depth length Z. Here, the horizontal direction, the vertical direction, and the depth direction may be orthogonal to each other. Also, the water surface may be defined as the upper surface of the space in the water. However, the above is an example, and the three-dimensional shape of the space in the water is not limited to a rectangular parallelepiped, and may be a cube, a cylindrical shape, etc.

[0020] Also, the echo image is called an echogram and is an image generated based on the sound pressure of the sound wave received as a reflected wave when the sound wave (transmission wave) transmitted from the water surface is reflected by fish or the like in a fish finder or the like. It is an image generated based on the sound pressure of the sound wave received as a reflected wave when the sound wave (transmission wave) transmitted from the water surface is reflected by fish or the like.

[0021] The size of the cut-out window is preferably determined such that the product of the estimation accuracy of a predetermined number of individuals and the required number of individuals is 1 or more. Also, the size of the cut-out window is preferably calculated based on the depth width of the reflected wave of the sound wave, the transmission period and the number of times of the sound wave. Also, it is preferable that the size of the cut-out window is further calculated based on the transmission beam width and the reception beam width of the sound wave.

[0022] The simulation of the behavior of the target fish calculates the position of each fish existing in the space in the water (fish behavior, that is, the time change of the position of each fish) based on the equation showing the behavior of the target fish in the space in the water (the equation of motion of the fish), the size of the space in the water (such as a fish trap), etc. The simulation may be a numerical simulation or otherwise. The equation of motion of the fish can be appropriately set according to the habits of the target fish.

[0023] The simulation of the behavior of target fish may simulate the behavior (position) of each fish existing in the underwater space (fish pen) based on at least one selected from, for example, attractive force, alignment force, repulsive force, propulsion force, water resistance, light repulsion, or pressure from a wall (boundary). The number of fish is arbitrary, but for example, it may be 100 to 10,000. Also, different sizes (body lengths) may be given to each fish. Since it is known that the body length distribution of fish in an actual fish pen follows a normal distribution, the body length of the fish may be given so that the distribution of the body length of each fish follows a normal distribution.

[0024] For example, assume that there are a plurality of fish in an underwater space such as a fish pen installed in the ocean or the like. A net or the like is installed at the boundary between the inside and the outside of the fish pen so that the fish cannot move back and forth between the inside and the outside of the fish pen. The size of the fish pen can be defined, for example, by the length in the horizontal (x) direction, the length in the vertical (y) direction, and the depth (length in the z direction) that are orthogonal to each other. The fish, for example, swim in a circular motion in a certain direction in the fish pen in a group.

[0025] The fish in the group try to approach each other. The attractive force can be expressed as the force acting between two individuals (two fish) that try to approach each other. Also, the fish try to avoid colliding with each other. The repulsive force can be expressed as the force acting between two individuals that try to avoid a collision. Also, the fish in the group try to suppress energy consumption by riding on the water flow created by the surrounding fish or move towards a place with a common purpose such as food. At this time, the fish try to match its speed with other surrounding fish. What represents this as the force acting between two individuals is the alignment force. The swimming force that the fish itself has is called the propulsion force, and the force received from the water against its movement is called the resistance force.

[0026] It is known that depending on the type of fish, many fish exhibit phototropism, which is the tendency to move away from a light source. Due to phototropism, fish may concentrate below the fishing net. Also, fish tend to gather on the bottom depending on the conditions of phototropism, but it is thought that their speed will decrease if they get too close to the bottom. Therefore, the speed of the fish may be reduced according to the distance from the bottom. Also, when the actual position of the fish exceeds the bottom surface, for example, it may be moved to a position that serves as a reference from the bottom surface.

[0027] Also, since the inside of the fishing net is dense, the outermost fish always experiences a force that presses it against the wall from the neighboring individuals, but the outermost fish stays put without colliding with the wall. This state can also be said to be receiving a force (pressure from the wall) directed from the wall towards the center of the fishing net. Next, consider the fish in the second position from the wall. If this fish also tries to move towards the wall, since there is a first fish from the wall located between this fish and the wall, it has no choice but to be stopped by a force directed towards the center opposite to the wall. In other words, the second fish is also receiving a force directed towards the center from a pseudo-wall. However, this pseudo-wall (the first fish from the wall) has a slightly movable range compared to the actual wall, so it is thought that the force directed towards the center received by the second fish is weaker than the force received by the first fish. Similarly, it is thought that the fish in the third and subsequent positions from the wall are also receiving a gradually weakening force directed towards the center. Therefore, it can be considered that a stronger force directed towards the center of the fishing net acts closer to the wall (farther from the center). Moreover, a viewing angle or a turnable angle may be set for the fish. Noise is given, for example, to follow four normal distributions with an average of 0 and standard deviations of 0.0, 0.02, 0.1, and 1.0. When the average is 0 and the standard deviation is 0.0, there is no noise. Depending on the type of fish, an intraspecific ranking relationship affected by body size has developed, and it is known to show aggressive behavior towards the same species around the feeding area. Such behavior may be considered.

[0028]

[0029] ​Hereinafter, the configuration of the embodiment will be further described with reference to the drawings. The configuration of the embodiment is an example. The configuration of the present disclosure is not limited to the configuration of the embodiment. The configuration of the embodiment can be appropriately changed without departing from the object of the present disclosure.

[0030] <Information processing apparatus> FIG. 1 is a diagram showing a configuration example of an information processing apparatus. In FIG. 1, the information processing apparatus 10 By executing a computer program, it can operate as a simulated echo image generation apparatus, an estimator generation apparatus, and an estimator of the number of fish individuals.

[0031] The information processing apparatus 10 is a dedicated or general-purpose computer such as a personal computer (PC), a workstation (WS), a server machine, or a smart device (such as a smartphone or a tablet terminal).

[0032] The information processing apparatus 10 includes a processor 101, a storage device 102, an input device 103, a display 104, and a communication interface (communication I / F) 105 that are interconnected via a bus B.

[0033] The storage device 102 is a non-temporary storage medium (computer-readable medium) capable of storing a computer program and data. The storage device 102 includes a main storage device (memory) and an auxiliary storage device. The main storage device (memory) is, for example, a RAM (Random Access Memory), or a RAM and a ROM (Read Only Memory), etc.

[0034] The auxiliary storage device is, for example, an EEPROM (Electrically Erasable Programmable ROM), a hard disk drive (HDD), an SSD (Solid State Drive), or a flash memory, etc.

[0035] The input device 103 is a key, a button, or a pointing device (such as a mouse), and is used for inputting information and data. The display 104 outputs (displays) information and data.

[0036] The communication I / F 105 is connected to other devices (such as a computer) through a wired or wireless network, and controls the processing related to communication with other devices (communication partners). For example, the information processing device 10 can communicate with the fish finder 20 through the communication I / F 105.

[0037] The processor 101 is composed of at least one of a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a DSP (Digital Signal Processor). By executing the program stored in the storage device 102, the processor 101 operates the information processing device 10 as a simulated echo image generation device, an estimator generation device, and an estimator for the number of fish individuals.

[0038] The processing performed by the processor 101 may be performed using a semiconductor device such as an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a CPLD (Complex Programmable Logic Device), or hardware such as an SoC (System on a Chip). The processor 101 and the above-described hardware are an example of circuitry that performs a desired process or operation.

[0039] The processor 101 simulates the behavior of a predetermined number of target fish (tuna) swimming within a virtual net pen by executing a program. Also, the processor 101 performs a simulation (acoustic simulation) when transmitting a sound wave into the virtual net pen and receiving a reflected wave from the target fish by executing a program, and can generate a simulated echo image, that is, a simulated echo image based on the acoustic simulation result. That is, the information processing apparatus 10 can generate a simulated echo image that simulates the reflected wave when a sound wave is transmitted into a virtual net pen in a simulation of the behavior of a predetermined number of fish (target fish) moving according to the equation of motion within the virtual net pen (an example of a space in a predetermined body of water). Further, the processor 101 can also generate an estimator using the simulated echo image and estimate (predict) the number of fish using the estimator.

[0040] <Generation of Simulated Echo Image> FIG. 2 is a flowchart showing an example of the generation process of a simulated echo image in the information processing apparatus 10. In step S01, the processor 101 calculates the required number of individuals, which is the number of fish that need to be present within the simulated echo image (cutout window), based on the estimation accuracy of the number of target fish (an example of a predetermined number of individuals) within the virtual net pen.

[0041] For example, when the number of target fish (total number) present within the virtual net pen is A [tails] and the estimation accuracy of the number is B [%], the required number C [tails] can be obtained as follows. That is, letting the estimation error with respect to the total number A of target fish within the net pen be a [tails], the estimation error a, for example, when the estimation error is 10 tails, is a = 10.

[0042] The estimation accuracy B can be obtained using the formula B = a÷A×100, or a = A×B / 100. And as the required number C, C that satisfies the condition C×B / 100≧1 is obtained. It follows that. Substituting B = a÷A×100 into the above formula, we get C×a / A ≧ 1, and C ≧ A / a. Thus, the required number of individuals C is determined such that the product of the predetermined estimation accuracy B and the required number of individuals C is 1 or more.

[0043] In step S02, the processor 101 determines the size of the cut-out window so that there are at least C target fish, where C is the required number of individuals obtained by dividing the total number A of target fish by the estimation error a. In step S03, the processor 101 performs a simulation of the behavior of the target fish in the virtual net pen and an acoustic simulation of the net pen, and generates a simulated echo image (cut-out image) of the size of the cut-out window in which at least C target fish are shown (see FIGS. 7A and 7B).

[0044] <Generation of estimator> In step S04, the processor 101 performs machine learning using the data of the cut-out image. For example, machine learning is performed using the data of the cut-out image as training data. By machine learning, an estimator for estimating the total number of target fish in the net pen is generated. Also, the data of the cut-out image may be used as test data. Further, in machine learning, validation data may be used. The information processing apparatus 10 that performs the process of step S104 operates as an estimator generation apparatus.

[0045] The process (machine learning) in step S104 may be a process independent of the processes in steps S01 to S03. That is, the information processing apparatus 10 can operate as an apparatus that performs machine learning using the data of the cut-out image as training data or test data, that is, an apparatus (estimator generation apparatus) that generates an estimator for estimating the total number of target fish in the net pen. In this case, the data of the cut-out image may be obtained by the information processing apparatus 10 operating as a simulated echo image generation apparatus, or the information processing apparatus 10 may acquire the data of the cut-out image generated by an apparatus or device other than the information processing apparatus 10 using the input device 103 or the communication I / F 105.

[0046] <Estimation of the number of fish> FIG. 3 is a flowchart showing an example of a process for estimating the total number of target fish existing in an actual (real) fish pond, which is performed by the information processing apparatus 10. The process shown in FIG. 3 shows an example of a process when the information processing apparatus 10 operates as an estimator for estimating the number of fish individuals. The information processing apparatus 10 has been trained by machine learning and is in a state where it can operate as an estimator. In step S11, the processor 101 acquires an echo image from the fish finder 20.

[0047] The fish finder 20 is set in an actual (real) fish pond having the same size and the same environment as the virtual fish pond. The fish finder 20 has a transducer (transmitting and receiving wave device) that transmits sound waves in a vertical direction from a predetermined position on the water surface of the fish pond into the water and receives reflected waves from target fish in the fish pond. Note that the direction of the sound wave beam is not necessarily the vertical direction and may be a non-vertical direction such as an oblique direction. The fish finder 20 can generate data of an echo image based on the received reflected waves and transmit it to the information processing apparatus 10. The data of the echo image is received by the information processing apparatus 10 through the network and the communication I / F 15 and stored in the storage device 102.

[0048] FIGS. 4 and 5 show examples of the installation positions (equipment positions) of the virtual and real fish ponds 30 and the fish finder 20. The left diagrams in FIGS. 4 and 5 are views of the fish pond 30 seen from the xy plane of the xyz orthogonal coordinate system, and the right diagrams are views of the fish pond 30 seen from the xz plane.

[0049] The fish basket 30 is a net formed in a cylindrical shape with an xy plane that is circular with a radius of 20 m and is installed in the sea (an example of open water), and its lower part is bowl-shaped. The central axis of the fish basket 30 (the z-axis passing through the origin (0, 0) of the xy-axis) passes through its deepest part, and the depth of the deepest part is about 20 m. The shape of the fish basket 30 is an example, and its shape and size are not limited to this example. The transducer of the fish finder 20 is installed at the position of the water surface (x, y) = (10, 0) which is 10 m away from the center (x, y) = (0, 0) of the fish basket 30 in the x direction, and it can emit sound waves vertically from the water surface. Emitting (transmitting) sound waves is called a ping, and the period of emitting (transmitting) a ping is called the ping rate. Note that each individual sound wave emitted (transmitted) is called a "pulse" or a "transmission pulse".

[0050] In FIGS. 4 and 5, the target fish (tuna) is indicated by a point, and in the fish basket 30, for example, it swims counterclockwise around the central axis of the fish basket 30. The number of individuals (total number A) of the target fish is 850. FIG. 4 shows the case of transmitting a sound wave with a frequency of 50 kHz and a transmission / reception beam width of 27° (-6 dB). FIG. 5 shows the case of transmitting a sound wave with a frequency of 200 kHz and a transmission / reception beam width of 9° (-6 dB). In FIGS. 4 and 5, when a sound wave is transmitted once, the fish bodies that generate a reflected wave of 100 dB (1 μPa / m) or more are illustrated as circles larger than the points.

[0051] The fish finder 20 needs to generate echo image data using the same parameters as those of a virtual fish finder set in an acoustic simulation for generating a simulated echo image. The parameters are, for example, the frequency of the transmitted and received waves, the ping rate (the transmission period of the sound wave (ping)), the number of transmissions of the ping, the transmission pulse length, the beam width (also called the beam angle (direction angle)), the detection depth width (depth range width), the transmission level, the reception sensitivity, etc. FIG. 5 shows the case of transmitting a sound wave with a frequency of 200 kHz and a transmission / reception beam width of 9° (-6 dB).

[0052] Regarding the depth range, the fish finder 20 needs to match the detection depth width with the depth width used when generating the simulated echo image. Also, the information processing device 10 synchronizes with the fish finder 20 in terms of the transmission period and number (time width) of sound waves (pings). In this way, the sizes of the simulated echo image and the actual echo image are aligned in the vertical (longitudinal) and horizontal (lateral) directions.

[0053] Returning to FIG. 3, in step S12, the processor 101 cuts out a portion from the echo image of the fish finder 20 corresponding to the cutout window of the simulated echo image. The necessary conditions for aligning the simulated echo image and the actual echo image are as follows. (1) Align the depth range width, which is the vertical axis of the image. (2) Align the time width (transmission period and number of sound waves), which is the horizontal axis of the image. (3) Align the beam width (directivity angle) of the transmitted and received waves. (4) Align the transmission level (power). (5) Align the pulse width. (6) Align the reception sensitivity. (7) Adjust the target strength (Ts, also called acoustic size) of the echo simulation to match the Ts of the tuna, which is the target fish. Ts is a measure of the detection target of the fish finder 20. (8) Align the transmission and reception frequencies.

[0054] FIG. 6 is a diagram showing examples of the simulated echo image and the actual echo image. The left diagram in FIG. 6 is an example of the simulated echo image, and the right diagram is an example of the echo image. The vertical axis of each image is the depth range width, indicated by the pixel value. The horizontal axis indicates the transmission period and number of sound waves (number of pings according to the ping rate, i.e., time width).

[0055] FIG. 7A is a diagram for explaining a cutout from a simulated echo image obtained by the virtual fish school detector 20 described with reference to FIG. 4, and FIG. 7B is a diagram for explaining a cutout from a simulated echo image obtained by the virtual fish school detector 20 described with reference to FIG. 5. The frequency of the sound wave used for creating the image shown in FIG. 7A is 50 kHz, and the frequency of the sound wave used for creating the image shown in FIG. 7B is 200 kHz. An example of the depth range width which is the vertical axis of the cutout window is 5.39 m to 20.57 m. Also, the horizontal axis (time width) of the cutout window is, for example, in units of 10 seconds (ping rate = 10 seconds), and as the time width becomes larger such as 10 seconds, 20 seconds, and 30 seconds, the horizontal width of the cutout window becomes larger.

[0056] Returning to FIG. 3, in step S14, the processor 101 outputs an estimated value of the number of fish. The estimated value is, for example, displayed on the display 104. The estimated value may be printed on a sheet by a printer.

[0057] <Simulation regarding simulated echo image generation> Hereinafter, calculation examples regarding simulated echo image generation and the number of fish body echoes in the echo image are shown.

[0058] (1. Fish school swimming situation and sound wave transmission / reception beam width) FIGS. 4 and 5 show simulation cases of the instantaneous value of the reflected wave intensity (one sound wave transmission and reflected wave observation) when sound wave irradiation and echo observation are performed on a fish school swimming in the live fish basket 30 using a fish school detector 20 (transducer) with different transmission / reception beam widths.

[0059] The simulation conditions are as follows. · In each of FIGS. 4 and 5, it is assumed that 850 tuna are swimming. · The installation position of the transducer of the fish school detector 20 is at the water surface (x = 10 m, y = 0 m) 10 m away in the radial direction from the center (x, y) = (0, 0) of the live fish basket 30. · In the simulated echo image generated, the echo signal that can be effectively used for estimator machine learning is one with a reflected received sound pressure of 100 dB or more (considering the relationship with background noise). In FIGS. 4 and 5, the fish bodies that generate reflected waves of 100 dB or more are indicated by circles larger than dots.

[0060] As shown in FIGS. 4 and 5, since the number of fish body reflected echoes observed varies depending on the beam width used, it is necessary to change the observation time according to the beam width to capture a predetermined number of reflected echoes.

[0061] (2. Relationship between the number of tails to be estimated, the observation position, and the observation time) FIG. 8 shows a calculation example of the number of fish body reflected echoes (those with a reflected received sound pressure of 100 dB or more) recorded in the simulated echo image when the number of tails in the live fish basket 30 and the installation position of the fish finder 20 are changed.

[0062] The simulation conditions are as follows. · Assume that 700 tuna are swimming in the live fish basket 30. · The installation positions of the fish finder 20 are at positions of 8 m, 9 m, 10 m, 11 m, 12 m, and 13 m in the radial direction (for example, the +X direction) from the center (0, 0) of the live fish basket. · The number of fish body echoes is aggregated from the reflected echoes in the depth range of 5.39 m to 20.57 m. · For cases where the time width of the simulated echo image is 10 seconds, 20 seconds, and 30 seconds, calculations were performed for 8 images, and the mean and variance were evaluated.

[0063] FIG. 8 shows the evaluation results in the case where the number of tuna tails is 700, FIG. 9 shows the evaluation results in the case where the number of tuna tails is 850, and FIG. 10 shows the evaluation results in the case where the number of tuna tails is 1000.

[0064] For example, in the case of the observation time of 10 seconds at the observation points of 8 m and 13 m shown in FIG. 8, the simulated echoes do not satisfy "determining the required number of individuals C such that the product of a predetermined estimation accuracy B (for example, B = 1%) and the required number of individuals C is 1 or more", and it can be seen that it is necessary to extend the observation time.

[0065] The calculation condition for the required number of individuals C, "the product of the predetermined estimation accuracy B and the required number of individuals C is 1 or more", is the minimum condition (sufficient condition for estimation resolution) for satisfying the predetermined estimation accuracy B. In reality, the number of individuals captured in each simulated echo image is affected by random fluctuations and has variations. The variation range was estimated and evaluated as the value of (5) in each table shown in FIGS. 8, 9, and 10. Assuming that the number of fish captured follows a normal distribution, it can be seen that an observation time of 30 seconds is required for the observation case at the 10 m point to achieve a result with an estimation accuracy of about ±2% with a probability of 95%.

[0066] <Experimental Example> FIGS. 11A and 11B are diagrams showing echo images (actual observation images) obtained using the actual fish finder 20. FIG. 11A shows the echo image of the observation in the first case, and FIG. 11B shows the echo image of the second case. These echo images were used as test data. [Data] · Cropping size: Vertical 518 pixels × Horizontal 160 pixels · Cropping range (vertical): Among 1020 pixels, the portion from 184 to 702 pixels (set based on the echo image in the first case) · Cropping range (horizontal): 10 seconds · Frequency of the fish finder: 50 kHz · Range of the fish finder (vertical resolution 1024 pixels): 20 m in the first case, and 30 m in the second case · Number of machine learning images: 993 learning data images (including 843 training data images and 149 validation data images), 200 test data images [Prediction Results] (First Case) The data used in the experiment and its prediction accuracy were as follows. · Learning data: Random noise was added in the range of 0 to 255 · Test data: Cropped as it is · Prediction accuracy: For the actual number of tails being 814, it was predicted (estimated) to be 855 tails. The error was 41 tails, and the error rate was approximately 5%. Figure 12A shows the results of the first case. The lower straight line in the graph indicates the actual number of tails, and the upper plotted points indicate the number of tails estimated using the estimator. The vertical axis of the graphs in Figures 12A and 12B is the number of fish, and the horizontal axis is the count number of the test data (since there are 200 pieces of test data, the values range from 0 to 200).

[0067] (The second case) The data used in the experiment and its prediction accuracy were as follows. · Training data: Cut out as it is · Test data: Cut out as it is · Prediction accuracy: For the actual number of tails being 908, it was predicted (estimated) to be 888 tails. The error was -20 tails, and the error rate was approximately -2.2%. Figure 12B shows the results of the second case. The upper straight line in the graph indicates the actual number of tails, and the lower plotted points indicate the number of tails estimated using the estimator.

[0068] Figure 13 is a table showing the number of fish with a reflected received sound pressure of 100 dB or more when a sound wave with a frequency of 50 kHz is emitted from the point (x, y) = (11, 0) in the live fish basket 30 shown in Figure 4. The observation time width is 10 seconds, and the depth width is 5.39 - 20.57 m.

[0069] For example, when the total number of target fish is 800 and the estimation error a = 10, the estimation accuracy is 10÷800×100 = 1.25%. And the required number of individuals C is such that C×1.25%≥1, so C = 80, and it is sufficient if 80 or more target fish are shown in the image. Also, when the total number of target fish is 900, C = 90, and it is sufficient if 90 or more target fish are shown in the image. In the table shown in Figure 13, for 800 and 900 fish respectively, the average number of fish far exceeds 80 and 90 respectively, and it can be seen that with the cut-out sizes of the above-mentioned depth width and time width, it is possible to estimate the number of fish with the desired estimation accuracy.

[0070] <The effects of the embodiment> According to an embodiment, a cut-out image (simulated echo image) with a required number of individuals C is generated as learning data for machine learning. Thereby, teacher data with a desired estimation accuracy can be generated. Further, an estimator can be generated by learning such teacher data. Then, an image that matches the conditions of the simulated echo image (cut-out image) in the estimator is cut out from the actual echo image and used as a comparison target with the simulated echo image, so that an accurate estimation result can be obtained.

Explanation of Signs

[0071] 10... Information processing device, 20... Fish school detector, 101... Processor, 102... Storage device

Claims

1. In a simulation where a predetermined number of fish move according to the equation of motion within a predetermined underwater space, a method for generating a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space, comprising: determining the size of the cutout window of the simulated echo image based on the required number of fish, which is the number of fish that need to exist in the simulated echo image, calculated based on the estimation accuracy of the predetermined number of fish; generating the simulated echo image with the determined size of the cutout window; A method for generating a simulated echo image including the above.

2. The size of the cutout window is determined such that the product of the estimation accuracy of the predetermined number of fish and the required number of fish is 1 or more. The method for generating a simulated echo image according to Claim 1.

3. The size of the cutout window is calculated based on the detection depth width of the reflected wave of the sound wave, the transmission period and the number of transmissions of the sound wave. The method for generating a simulated echo image according to Claim 1 or 2.

4. The size of the cutout window is further calculated based on the transmission beam width and the reception beam width of the sound wave. The method for generating a simulated echo image according to Claim 3.

5. In a simulation where a predetermined number of fish move according to the equation of motion within a predetermined underwater space, an apparatus for generating a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space, comprising: determining the size of the cutout window of the simulated echo image based on the required number of fish, which is the number of fish that need to exist in the simulated echo image, calculated based on the estimation accuracy of the predetermined number of fish; generating the simulated echo image with the determined size of the cutout window; a circuit that executes the above; An apparatus for generating a simulated echo image including the above.

6. In a simulation where a predetermined number of fish move according to the equation of motion within a predetermined underwater space, a program for causing a computer to generate a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space, comprising: determining the size of the cutout window of the simulated echo image based on the required number of fish, which is the number of fish that need to exist in the simulated echo image, calculated based on the estimation accuracy of the predetermined number of fish; generating the simulated echo image with the determined size of the cutout window; A program for causing the computer to execute the above.

7. A method for generating the estimator of the predetermined number of individuals using a simulation in which a predetermined number of fish move in a predetermined underwater space according to the equation of motion, Obtaining a simulated echo image generated with a cutout window size determined based on the required number of individuals, which is the number of fish that need to be present in the simulated echo image simulating the reflected wave when transmitting a sound wave into the underwater space, calculated based on the estimation accuracy of the predetermined number of individuals, Generating the estimator by learning using the simulated echo image, An estimator generation method including the above steps.

8. The size of the cutout window is determined such that the product of the estimation accuracy of the predetermined number of individuals and the required number of individuals is 1 or more. The estimator generation method according to claim 7.

9. The size of the cutout window is calculated based on the detection depth width of the reflected wave of the sound wave, the transmission period and the number of transmissions of the sound wave. The estimator generation method according to claim 7 or 8.

10. The size of the cutout window is further calculated based on the transmission beam width and the reception beam width of the sound wave. The estimator generation method according to claim 7.

11. An apparatus for generating the estimator of the predetermined number of individuals using a simulation in which a predetermined number of fish move in a predetermined underwater space according to the equation of motion, Obtaining a simulated echo image generated with a cutout window size determined based on the required number of individuals, which is the number of fish that need to be present in the simulated echo image simulating the reflected wave when transmitting a sound wave into the underwater space, calculated based on the estimation accuracy of the predetermined number of individuals, Generating the estimator by learning using the simulated echo image, A circuit for performing the above steps, An estimator generation apparatus including the above circuit.

12. An apparatus for causing a computer to generate the estimator of the predetermined number of individuals using a simulation in which a predetermined number of fish move in a predetermined underwater space according to the equation of motion, A step of obtaining a simulated echo image generated with a cutout window size determined based on the required number of individuals, which is the number of fish that need to be present in the simulated echo image simulating the reflected wave when transmitting a sound wave into the underwater space, calculated based on the estimation accuracy of the predetermined number of individuals, A step of generating the estimator by learning using the simulated echo image, A program for causing the computer to execute the above steps.

13. Obtaining an echo image showing a reflected wave when a sound wave is transmitted into a predetermined underwater space; estimating the number of fish present in the underwater space using the echo image and an estimator generated by learning a simulated echo image, including performing; The simulated echo image is a necessary number of fish that need to be present in a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space in a simulation where a predetermined number of fish move according to the equation of motion in the predetermined underwater space. The necessary number is determined based on the necessary number calculated based on the estimation accuracy of the predetermined number, and is generated with the size of a cutout window; The echo image is generated with the size of the cutout window; A method for estimating the number of fish.

14. The size of the cutout window is determined such that the product of the estimation accuracy of the predetermined number and the necessary number is 1 or more; The estimation method according to claim 13.

15. The size of the cutout window is calculated based on the detection depth width of the reflected wave of the sound wave, the transmission period and the number of transmissions of the sound wave; The estimation method according to claim 13 or 14.

16. Furthermore, the size of the cutout window is calculated based on the transmission beam width of the sound wave and the reception beam width of the reflected wave; The estimation method according to claim 15.

17. Obtaining an echo image showing a reflected wave when a sound wave is transmitted into a predetermined underwater space; A circuit including estimating the number of fish present in the underwater space using the echo image and an estimator generated by learning a simulated echo image; The simulated echo image is a necessary number of fish that need to be present in a simulated echo image that simulates the reflected wave when a sound wave is transmitted into the underwater space in a simulation where a predetermined number of fish move according to the equation of motion in the predetermined underwater space. The necessary number is determined based on the necessary number calculated based on the estimation accuracy of the predetermined number, and is generated with the size of a cutout window; The echo image is generated with the size of the cutout window; An apparatus for estimating the number of fish.

18. A step of obtaining an echo image showing a reflected wave when a sound wave is transmitted into a predetermined underwater space; A program that causes a computer to estimate the number of fish present in the underwater space by using the echo image and an estimator generated by learning a simulated echo image, wherein the simulated echo image is a required number of fish necessary to be present in a simulated echo image simulating a reflected wave when a sound wave is transmitted into the underwater space in a simulation in which a predetermined number of fish move according to the equation of motion within the predetermined underwater space, and is generated with a size of a cutout window determined based on the required number of fish calculated based on the estimation accuracy of the predetermined number of fish, the echo image is generated with the size of the cutout window Program.

Images