Fish count calculation method, fish count calculation program, and fish count calculation device

Patent metadata

Field	Value
Publication number	US-12532872-B2
Application number	US-202118031222-A
Country	US
Kind code	B2
Filing date	Oct 13, 2021
Priority date	Oct 14, 2020
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

In the present invention, an estimator for estimating the number of fish present in an underwater space is constructed by means of machine learning using, as teaching data, a plurality of data sets for learning that each include an echo image for learning, the echo image being based on received sound waves reflected by fish when sound waves are transmitted in an underwater space where fish are present, and the number of fish present in the underwater space in the echo image. The plurality of data sets for learning each include: an echo image for learning that is based on received sound waves reflected by fish when sound waves are transmitted in the underwater space where fish are present; and the number of fish present in the underwater space in the echo image. The estimator is used on an echo image generated on the basis of received sound waves reflected by an unknown number of fish present in the underwater space after transmitting sound waves in the underwater space, so as to calculate the number of the unknown number of fish present in the underwater space.

First claim

The invention claimed is:

1 . A fish count calculation method, comprising:
constructing, by a computer, an estimator for estimating a number of fish present in a pen through machine learning using, as training data, a plurality of training data sets, each of which includes a training echo image based on sound waves received upon being reflected by the fish traveling within the pen when sound waves are transmitted in the pen, and the number of fish present in the pen in the training echo image; and calculating, by the computer, an unknown number of fish present in a real pen by using the estimator on a real echo image generated based on sound waves that are transmitted in the real pen and received upon being reflected by the unknown number of fish present in the real pen, wherein a plurality of training echo images included in the plurality of training datasets include a simulated echo image generated using numerical simulation of fish behavior, the numerical simulation generating sound waves reflected from a predetermined number of fish traveling in a school within the pen.

2 . The fish count calculation method according to claim 1 , further comprising:
calculating, by the computer, using the numerical simulation, sound waves when the predetermined number of fish reflect sound waves transmitted from a simulated fish finder into the pen; and generating, by the computer, the simulated echo image based on echo sound pressure for each distance from the fish finder to the fish.

3 . The fish count calculation method according to claim 2 , further comprising: calculating, by the computer, in the numerical simulation, the behavior of the fish in the pen with a motion equation for each fish, by using, as parameters for the fish present in the pen, at least a size of the fish, a force acting on the fish, a field-of-view angle of the fish, a size of the pen, and a flow speed of a fluid in the pen.

4 . The fish count calculation method according to claim 3 , wherein the force acting on the fish includes an alignment force that acts between two individual fish within the school.

5 . The fish count calculation method according to claim 3 , wherein the force acting on the fish includes a downward force based on phototropism causing the fish to move away from a light source.

6 . The fish count calculation method according to claim 3 , wherein the force acting on the fish includes a force directed toward a center of the pen, wherein the force directed toward a center of the pen increases as the fish approaches the wall of the pen.

7 . The fish count calculation method according to claim 3 , wherein in the numerical simulation, the predetermined number of fish travel around inside of the pen in a shape of a truncated cone whose radius increases toward the bottom.

8 . The fish count calculation method according to claim 1 , wherein:
the fish present in the pen are classified into two or more classes of fish sizes; the estimator has learned to be able to estimate the number of fish present in the pen for each class, through machine learning using the plurality of training data sets; and the computer uses the estimator to calculate the number of fish present in the real pen for each class, with respect to the real echo image.

9 . A non-transitory computer-readable storage medium storing a program for causing a computer to:
construct an estimator for estimating a number of fish present in a pen through machine learning using, as training data, a plurality of training data sets, each of which includes a training echo image based on sound waves received upon being reflected by the fish when sound waves are transmitted in the pen in which the fish travel within the pen, and the number of fish present in the pen in the training echo image; and calculate an unknown number of fish present in a real pen by using the estimator on a real echo image generated based on sound waves that are transmitted in the real pen and received upon being reflected by the unknown number of fish present in the real pen, wherein a plurality of training echo images included in the plurality of training datasets include a simulated echo image generated using numerical simulation of fish behavior, the numerical simulation generating sound waves reflected from a predetermined number of fish traveling in a school within the pen.

10 . A fish count calculation apparatus, comprising:
a processor configured to: construct an estimator for estimating a number of fish present in a pen through machine learning using, as training data, a plurality of training data sets, each of which includes a training echo image based on sound waves received upon being reflected by the fish when sound waves are transmitted in the pen in which the fish travel within the pen, and the number of fish present in the pen in the training echo image; and calculate an unknown number of fish present in a real pen by using the estimator on a real echo image generated based on sound waves that are transmitted in the real pen and received upon being reflected by the unknown number of fish present in the real pen, wherein a plurality of training echo images included in the plurality of training datasets include a simulated echo image generated using numerical simulation of fish behavior, the numerical simulation generating sound waves reflected from a predetermined number of fish traveling in a school within the pen.

Assignees

• Japan Agency Marine Earth Sci

Inventors

• Nishikawa Haruka
• Ishikawa Yoichi
• Kuwatani Tatsu
• Matsuoka Daisuke
• Sugiyama Daisuke
• Watanabe Yoshitaka

Classifications

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T7/75

  involving models · CPC title

• G01S15/96

  for locating fish · CPC title

• A01K61/95 Primary

  specially adapted for fish · CPC title

• G01S15/89

  for mapping or imaging · CPC title