Fishery harvest prediction system and fishery harvest prediction method

Abstract

To precisely predict a fishery harvest.SOLUTION: A fishery harvest prediction system 10 is configured to predict a fishery harvest of fish and shellfish of previously set kinds, and comprises an acquisition part 11 which acquires information for prediction associated with a sea area of depth where the fish and shellfish inhabit, and a prediction part 12 which predicts a fishery harvest of the fish and shellfish based upon the information for prediction acquired by the acquisition part 11.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a fish catch prediction system and a fish catch prediction method. [Background technology]

[0002] In fixed net fishing, it is expected that the efficiency of fishing can be improved by predicting the catch volume. As a similar prediction, it has been proposed to predict the migration period of specific fish species using satellite data (for example, see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Satellite-based information service for fixed-net fishing, [online], Collection of good practice cases using satellite data in business, March 2020, pages 14-15, [search date: August 1, 2023],<URL:https: / / www8.cao.go.jp / space / goodpractice / r02 / r02_jirei_all.pdf> Summary of the Invention [Problem to be solved by the invention]

[0004] The conventional prediction shown in Non-Patent Document 1 uses satellite data for prediction. However, Non-Patent Document 1 does not specifically indicate what kind of information the satellite data is, and there is a risk that the data is not necessarily suitable for predicting fish catches. Therefore, the conventional prediction shown in Non-Patent Document 1 cannot necessarily predict fish catches with high accuracy.

[0005] The present invention has been made in consideration of the above, and has an object to provide a catch prediction system and a catch prediction method that can predict catch amounts with high accuracy. [Means for solving the problem]

[0006] In order to achieve the above-mentioned objective, the fish catch prediction system of the present invention is a fish catch prediction system that predicts the catch of a predetermined type of seafood, and is equipped with an acquisition means that acquires prediction information related to the water area at the depth where the seafood lives, and a prediction means that predicts the catch of the seafood based on the prediction information acquired by the acquisition means.

[0007] The prediction information used in the fish catch prediction system according to the present invention is suitable for predicting the catch of a predetermined type of fish and shellfish. Therefore, the fish catch prediction system according to the present invention can predict the catch with high accuracy.

[0008] The acquisition means may acquire information indicating a temperature of the water area as the prediction information. The acquisition means may acquire information indicating a water flow of the water area as the prediction information. With this configuration, it is possible to more reliably and accurately predict the catch amount.

[0009] The acquisition means may acquire prediction information for each of a plurality of positions in a water area in an area including a position where the catch is to be predicted. With this configuration, it is possible to more accurately predict the catch by taking into account the conditions of each of the plurality of positions in the area.

[0010] The prediction means may input the prediction information, output information indicating the catch amount, and predict the catch amount of fish and shellfish using a prediction model generated by machine learning. With this configuration, it is possible to more reliably and accurately predict the catch amount.

[0011] Incidentally, the present invention can be described not only as an invention of a fish catch prediction system as described above, but also as an invention of a fish catch prediction method as described below. These are essentially the same inventions, with the only difference being in different categories, and they have similar functions and effects.

[0012] In other words, the catch prediction method of the present invention is a catch prediction method that is an operating method of a catch prediction system that predicts the catch of a predetermined type of seafood, and includes an acquisition step of acquiring prediction information related to the water area at the depth where the seafood lives, and a prediction step of predicting the catch of the seafood based on the prediction information acquired in the acquisition step. Effect of the Invention

[0013] According to the present invention, the catch volume can be predicted with high accuracy. [Brief description of the drawings]

[0014] [Figure 1] 1 is a diagram showing a functional configuration of a fish catch prediction system according to an embodiment of the present invention. [Diagram 2] FIG. 13 is a diagram showing an example of prediction information used to predict a catch volume. [Diagram 3] 1 is a graph showing an example of the catch of Akisake salmon. [Figure 4] 1 is a graph showing an example of yellowtail catch volume. [Diagram 5] 1 is a graph showing an example of the catch volume of Japanese flying squid. [Figure 6] 4 is a flowchart showing a fish catch prediction method which is a process executed in the fish catch prediction system according to the embodiment of the present invention. [Figure 7] FIG. 11 is a scatter plot of an example of an analysis result of the prediction results. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment of a fish catch prediction system and a fish catch prediction method according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are given the same reference numerals and duplicated explanations will be omitted.

[0016] Fig. 1 shows the functional configuration of a fish catch prediction system 10 according to this embodiment. The fish catch prediction system 10 is a system (device) that predicts the catch of a preset type of fish and shellfish. The fish catch prediction system 10 predicts the catch in a preset fishing ground. The fishing ground is, for example, an ocean fishing ground. However, the fishing ground may be a fishing ground other than the ocean, such as a river or a lake.

[0017] The catch prediction system 10 predicts the catch in, for example, fixed net fishing. In this case, the type of seafood for which the catch is predicted is seafood that is the target of set net fishing, such as migratory fish. Specific types of seafood for which the catch is predicted are salmon, yellowtail, and Pacific pacific flying squid. Information indicating the predicted catch is provided to, for example, fishermen. Fishermen can fish more efficiently by referring to the information indicating the predicted catch. However, the fishing method related to the predicted catch may be a method other than set net fishing.

[0018] Specifically, the fish catch prediction system 10 is configured by a computer equipped with hardware such as a CPU (Central Processing Unit) and memory. Each function of the fish catch prediction system 10, which will be described later, is realized by these components operating through programs or the like. The fish catch prediction system 10 may be realized by one computer, or may be realized by a computer system configured by connecting a plurality of computers to each other via a network.

[0019] Next, a description will be given of functions of the fish catch prediction system 10 according to this embodiment. As shown in FIG.

[0020] The acquisition unit 11 is an acquisition means for acquiring prediction information related to a water area of ​​a depth where fish and shellfish of a species for which the catch is to be predicted live. The acquisition unit 11 may acquire information indicating the temperature of the water area as the prediction information. The acquisition unit 11 may acquire information indicating the water flow in the water area as the prediction information. The acquisition unit 11 may acquire prediction information for each of a plurality of positions in the water area in an area including a position for predicting the catch.

[0021] An example of the prediction information is shown as an image in Figure 2. The example of prediction information shown in Figure 2 is for estimating the catch volume in fishing grounds in Iwate Prefecture. The prediction information is information for each of a plurality of positions in the waters of an area that includes the position where the catch volume is predicted. For example, the area is divided horizontally (in the latitude and longitude directions) into a plurality of meshes, and the prediction information is information related to the waters of each mesh. In the example shown in Figure 2, the area includes the sea area around eastern Japan at latitude 35-47 degrees north and longitude 139-135 degrees east, as well as the Oyashio region and off the coast of the Pacific Ocean.

[0022] The prediction information is information about a water area at a time that has a specific relative time relationship to the time for which the catch is predicted. These times may be a period of time having a certain length, for example 10 days. Typically, the time related to the prediction information is the same as the time for which the catch is predicted or a time prior to the time for which the catch is predicted.

[0023] The prediction information is, for example, a physical quantity related to the water area. For example, the prediction information is information indicating the temperature (water temperature) of the water area, or information indicating the water flow (e.g., ocean currents, tidal currents) of the water area. When the time related to the prediction information is a time period having a certain length (e.g., 10 days), the prediction information may correspond to these time periods. For example, the prediction information may be the average value of these time periods.

[0024] The prediction information may be information measured in the water body, or information predicted by some method. The measurement or prediction may be performed by any method, including conventional methods.

[0025] The prediction information is information related to the water area at the depth inhabited by the fish and shellfish species for which the catch is to be predicted. For example, if the species for which the catch is to be predicted is salmon, yellowtail, or Japanese flying squid, the relevant depth is 50m-200m. Note that the prediction information does not need to relate to all of these depth ranges, but may relate to any of these depth ranges. The example of prediction information shown in FIG. 2 is used to predict the catch of the above-mentioned species of fish and shellfish.

[0026] The prediction information shown in Figure 2(a) is information indicating the temperature (water temperature) of the water area. The prediction information shown in Figure 2(a) is water temperature values ​​for multiple positions (meshes) at a depth of 100 m. The values ​​are averages over 10 days. The prediction information shown in Figure 2(b) is information indicating the water flow in the water area. The prediction information shown in Figure 2(b) is information indicating ocean currents for multiple positions (meshes) at a depth of 50 m. The information indicating the ocean currents is, for example, a vector indicating the direction and strength of the water flow. The vectors are averages over 10 days.

[0027] In this manner, the prediction information may be information on the distribution of water temperature or ocean current (for example, information that can be represented as an image as shown in FIG. 2). The prediction information may also include multiple types of information. For example, the prediction information may include information on both water temperature and ocean current.

[0028] As prediction information, it is conceivable to use satellite data (e.g., water temperature only at the sea surface or the color of the sea surface) that can be obtained by measurements from artificial satellites. However, when targeting fish and shellfish whose inhabiting depths are more than a certain distance from the water surface (e.g., fish and shellfish that are subject to fixed net fishing), information related to the water area depth cannot be generated from satellite data alone (or from satellite data itself; the same applies below). Therefore, when targeting such fish and shellfish, prediction information needs to be generated from data other than satellite data. Even if prediction information cannot be generated from satellite data alone, satellite data may be used in addition to data other than satellite data to generate the prediction information. Furthermore, when targeting fish and shellfish whose inhabiting depths are near the water surface, prediction information may be generated from satellite data alone.

[0029] As the prediction information, for example, ocean information published by a public institution can be used. An example of the prediction information shown in Fig. 2 is ocean information published by the Japan Meteorological Agency. By using the information published by a public institution, the manager and users of the fish catch prediction system 10 do not need to install sensors to obtain the prediction information themselves.

[0030] The acquiring unit 11 acquires the prediction information by, for example, accepting an input operation of the prediction information from a user. Alternatively, the acquiring unit 11 may acquire the prediction information from another device. Moreover, the acquiring unit 11 may acquire the prediction information by a method other than the above. The acquiring unit 11 outputs the acquired prediction information to the prediction unit 12.

[0031] The prediction unit 12 is a prediction means for predicting the catch of fish and shellfish based on the prediction information acquired by the acquisition unit 11. The prediction unit 12 may input the prediction information and output information indicating the catch, and may predict the catch of fish and shellfish using a prediction model generated by machine learning. The prediction unit 12 predicts the catch of fish and shellfish as follows.

[0032] The prediction unit 12 inputs the prediction information from the acquisition unit 11. The prediction unit 12 predicts the catch of seafood at a time having a specific relative time relationship with the time related to the input prediction information (for example, the time when a value related to the prediction information is measured). For example, the prediction unit 12 predicts the catch of seafood for a 10-day period related to the input prediction information. That is, in this case, the time (period) related to the input prediction information and the time (period) to be predicted are the same. The prediction unit 12 may also predict the catch of seafood for a period beyond the 10 days related to the input prediction information. The specific relative time relationship is determined in advance.

[0033] The prediction unit 12 predicts, for example, which of a plurality of preset classes the catch will fall into. The plurality of classes is, for example, a total of four classes, including a class for a catch of 0 kg and three classes obtained by dividing the area exceeding 0 kg into three.

[0034] Figures 3 to 5 show example graphs of catch volume. The graph in Figure 3 is a graph of the catch volume of Akisake salmon. The graph in Figure 4 is a graph of the catch volume of Yellowtail salmon. The graph in Figure 5 is a graph of the catch volume of Pacific pacific flying squid. All of these graphs are of the fixed net catch volume in Kuji City, Iwate Prefecture. In these graphs, the horizontal axis shows time, and the vertical axis shows the average catch volume [kg] in 10-day intervals.

[0035] These graphs also show four classifications. The dashed lines in the graphs indicate the catch amounts at the class boundaries. In the catch amounts of Akisake shown in Figure 3, class C1 is 0 kg, class C2 is more than 0 kg and less than 5000 kg, class C3 is 5000 kg to 20000 kg, and class C4 is more than 20000 kg. In the catch amounts of Yellowtail shown in Figure 4, class C1 is 0 kg, class C2 is more than 0 kg and less than 2000 kg, class C3 is 2000 kg to 5000 kg, and class C4 is more than 5000 kg. In the catch amounts of Pacific flying squid shown in Figure 5, class C1 is 0 kg, class C2 is more than 0 kg and less than 200 kg, class C3 is 200 kg to 1000 kg, and class C4 is more than 1000 kg.

[0036] The number of classes and the classification do not have to be as described above, and may be set according to the needs of fishermen or the type of fish and shellfish. Generally, the greater the number of classes, the lower the accuracy of the prediction. The prediction unit 12 may predict the catch as a value, for example, a value in kg, instead of the classes described above. The prediction unit 12 may also predict a result other than the above, so long as it is a prediction of the catch.

[0037] The prediction unit 12 stores in advance a prediction model for predicting the catch of fish and shellfish from the prediction information, and performs prediction using the prediction model. The prediction model is, for example, a trained model generated by machine learning. Specifically, the prediction model may include a convolutional neural network (CNN). More specifically, the prediction model may use ResNet-101, VGG19, or the like. As described above, if the prediction information can be handled as distribution information, that is, an image, appropriate prediction can be made by using CNN, which is generally used in the field of image recognition. In addition, the prediction model may predict the value of the catch by, for example, a regression model. In addition, the prediction model may be a trained model other than the above, and may be a model other than machine learning.

[0038] The prediction model receives the prediction information acquired by the acquisition unit 11 as input, and outputs information indicating the predicted result of the catch. The prediction model is configured to be able to receive the prediction information. For example, the input layer of the prediction model is provided with neurons equal to the number of pieces of prediction information data. The prediction model is also configured to output a prediction result corresponding to the prediction information. For example, the output layer of the prediction model is provided with neurons equal to the number of classes that output a preset score for each class. A higher score indicates a higher likelihood of the catch of that class. Alternatively, the output layer of the prediction model is provided with neurons that output a predicted value of the catch.

[0039] The prediction unit 12 inputs the prediction information input from the acquisition unit 11 into a prediction model, and obtains information indicating the prediction result of the catch, for example, the score for each class described above, as an output from the prediction model. The prediction unit 12 outputs the information indicating the prediction result of the catch. For example, the prediction unit 12 transmits the information indicating the prediction result of the catch to a fisherman's terminal. The information may be transmitted to the fisherman via a website or a smartphone application as an information distribution service for fishermen. The information indicating the prediction result from the prediction unit 12 may be output by a method other than the above and to an output destination other than the above.

[0040] A prediction model is generated, for example, as follows. First, learning data (teacher data) required for generating a prediction model is acquired. The learning data includes information to be input to the prediction model (i.e., information equivalent to prediction information) and the catch volume of seafood at a time having the above-mentioned specific relative time relationship with the time related to the information. In other words, the learning data is prediction information in which the catch volume is known. The amount of acquired learning data is sufficient for generating an appropriate prediction model. For example, data from the past 30 years may be used as learning data.

[0041] Next, machine learning is performed based on the learning data. Specifically, machine learning is performed in which information (i.e., information equivalent to prediction information) that is to be input to the prediction model corresponding to each other among the learning data is input to the prediction model, and information based on the catch is output to the prediction model. In the case of a prediction model that predicts a class for each catch, the information based on the catch that is output to the prediction model is, for example, information in which the value of the class including the catch is 1 and the value of the other classes is 0. In the case of a prediction model that predicts the catch itself, the information based on the catch that is output to the prediction model is the value of the catch. The training of the above machine learning itself, that is, the update of the parameters of the prediction model may be performed in the same manner as conventional machine learning training. Note that the generation of the prediction model by machine learning may be performed by the catch prediction system 10 or may be performed by a system other than the catch prediction system 10.

[0042] The prediction model, which is a trained model used in the fish catch prediction system 10, is expected to be used as a program module that is a part of artificial intelligence software. The model is used in a computer equipped with hardware such as a CPU and a memory, for example, and the CPU of the computer operates according to instructions from the model stored in the memory. For example, the CPU of the computer operates to input information to the model according to the instructions, perform calculations according to the model, and output results from the model. Specifically, the CPU of the computer operates to input information to the input layer of the neural network according to the instructions, perform calculations based on trained weighting coefficients in the neural network, and output results from the output layer of the neural network. The above is the function of the fish catch prediction system 10 according to this embodiment.

[0043] Next, a fish catch prediction method, which is a process executed by the fish catch prediction system 10 according to this embodiment (a method of operation performed by the fish catch prediction system 10), will be described with reference to the flowchart of FIG.

[0044] In this process, first, the acquisition unit 11 acquires prediction information relating to the water area of ​​the depth inhabited by the fish and shellfish for which the catch is to be predicted (S01, acquisition step). Next, the prediction unit 12 predicts the catch of the fish and shellfish based on the prediction information acquired by the acquisition unit 11 (S02, prediction step). Next, information showing the prediction result of the catch of the fish and shellfish is output from the prediction unit 12 (S03). The above is the catch prediction method which is the process executed by the catch prediction system 10 according to this embodiment.

[0045] The prediction information used in this embodiment is suitable for predicting the catch volume of a preset type of fish and shellfish. Therefore, this embodiment makes it possible to predict the catch volume with high accuracy. Accurate prediction of the catch volume can contribute to, for example, the efficiency and management stability of fixed net fisheries.

[0046] Furthermore, as in this embodiment, the prediction information may be information indicating the temperature of the water area. Furthermore, the prediction information may be information indicating the water flow in the water area. These configurations allow for more reliable and accurate prediction of the catch volume. However, the prediction information may be other than the above information as long as it relates to a water area with a depth where fish and shellfish live. Furthermore, the prediction information may include multiple types of information.

[0047] Furthermore, as in this embodiment, the prediction information may be information for each of a plurality of positions in the waters of an area including the position where the catch is predicted. For example, the prediction information may be information on the distribution of a wide area as shown in FIG. 2. With this configuration, the catch can be predicted more accurately by taking into account the conditions of each of a plurality of positions in the area. However, the prediction information does not have to be information for each of a plurality of positions.

[0048] As in the present embodiment, the catch may be predicted by inputting prediction information, outputting information indicating the catch, and using a prediction model generated by machine learning. This configuration allows the catch to be predicted more reliably and accurately. However, the catch does not necessarily have to be predicted using the prediction model, and may be predicted based on the prediction information.

[0049] The catch was predicted using a prediction model generated by machine learning using the catch data and oceanographic information from the past 30 years published on the Web. The catches were predicted for each of Akisake, Yellowtail, and Pacific flying squid in the fishing grounds of Iwate Prefecture. In this example, ResNet-101 was used as the prediction model. The prediction information used was the area around eastern Japan at 35-47 degrees north latitude and 139-135 degrees east longitude, including the Oyashio region and the Pacific coast, as shown in Figure 2. The prediction information included the water temperature at a depth of 100 m and information indicating the ocean current at a depth of 50 m. The prediction result was one of the four classes mentioned above.

[0050] When using water temperature at a depth of 100m as prediction information, the prediction accuracy rate was 78.74% for Akisake, 76.62% for Yellowtail, and 70.09% for Pacific flying squid. When using ocean currents at a depth of 50m as prediction information, the prediction accuracy rate was 79.68% for Akisake, 73.00% for Yellowtail, and 72.62% for Pacific flying squid. In all cases, the catch volume was predicted with an accuracy rate of 70-80%. As these results show, the catch volume was predicted with high accuracy. Note that the accuracy rate here is for predictions made using validation prediction information that was not used in machine learning.

[0051] As another example of prediction, out of multiple pieces of prediction information with known catch volumes, 80% was used for machine learning, and the remaining 20% ​​was used as data for validation to predict the catch volume. The catch volume prediction target was Akisake in fishing grounds in Iwate Prefecture. In this example, VGG19_BN was used as the prediction model. The prediction result is the catch volume value. Other than that, it is the same as the above example.

[0052] Figure 7(a) shows a scatter plot of the catch volume predicted using the prediction information used in machine learning (80% of the prediction information) and the actual catch volume corresponding to the prediction information (the catch volume used as a label). Figure 7(b) shows a scatter plot of the catch volume predicted using the prediction information for validation (20% of the prediction information) that was not used in machine learning and the actual catch volume corresponding to the prediction information. In the scatter plot in Figure 7, the horizontal axis shows the actual catch volume and the vertical axis shows the predicted catch volume. When the prediction information for validation shown in Figure 7(b) was used, the coefficient of determination for the regression line obtained from the actual catch volume and the predicted catch volume was 0.71.

[0053] The fish catch prediction system and fish catch prediction method disclosed herein have the following configuration. [1] A catch prediction system for predicting the catch of a predetermined type of fish or shellfish, comprising: An acquisition means for acquiring prediction information related to the water area where the fish and shellfish live; a prediction means for predicting the catch amount of the fish and shellfish based on the prediction information acquired by the acquisition means; A fish catch prediction system equipped with [2] The fish catch prediction system described in [1], wherein the acquisition means acquires information indicating the temperature of the water area as prediction information. [3] The fish catch prediction system according to [1] or [2], wherein the acquisition means acquires information indicating the water flow in the water area as prediction information. [4] A fish catch prediction system according to any one of [1] to [3], wherein the acquisition means acquires prediction information for each of a plurality of positions in the water area including a position where the catch is predicted. [5] A catch prediction system described in any of [1] to [4], wherein the prediction means inputs prediction information and outputs information indicating the catch, and predicts the catch of the seafood using a prediction model generated by machine learning. [6] A method for predicting a catch of a predetermined type of fish or shellfish, comprising: An acquisition step of acquiring prediction information related to the water area where the fish and shellfish live; a prediction step of predicting the catch amount of the fish and shellfish based on the prediction information acquired in the acquisition step; A method for forecasting fish catches, including: [Explanation of symbols]

[0054] 10...fish catch prediction system, 11...acquisition unit, 12...prediction unit.

Claims

1. A catch prediction system for predicting a catch of a predetermined type of fish and shellfish, An acquisition means for acquiring prediction information related to the water area where the fish and shellfish live; a prediction means for predicting the catch amount of the fish and shellfish based on the prediction information acquired by the acquisition means; A fish catch prediction system equipped with

2. The fish catch prediction system according to claim 1 , wherein the acquisition means acquires information indicating a temperature of the water area as prediction information.

3. The fish catch prediction system according to claim 1 or 2, wherein the acquisition means acquires information indicating the water flow of the water area as prediction information.

4. The fish catch prediction system according to claim 1 or 2, wherein the acquisition means acquires prediction information for each of a plurality of positions in the water area including a position where the catch is to be predicted.

5. The catch prediction system according to claim 1 or 2, wherein the prediction means inputs prediction information, outputs information indicating the catch, and predicts the catch of the seafood using a prediction model generated by machine learning.

6. A method for predicting a catch of a predetermined type of fish or shellfish, comprising the steps of: An acquisition step of acquiring prediction information related to the water area where the fish and shellfish live; a prediction step of predicting the catch amount of the fish and shellfish based on the prediction information acquired in the acquisition step; A method for forecasting fish catches, including:

Images