Detection apparatus and detection method

Abstract

A detection apparatus 1 for detecting an upwelling region includes: a simulation unit 11 for simulating sea surface temperature data, chlorophyll-a concentration data, and ocean current velocity data of regions in the ocean; a preprocessing unit 12 for creating a ground truth label of an upwelling region on the basis of a vertical velocity included in the ocean current velocity data; a classification unit 14 for training a learning model for outputting a classification result of an upwelling region in response to the input of sea surface temperature data and chlorophyll-a concentration data, wherein the training is performed by inputting the simulated sea surface temperature data and chlorophyll-a concentration data to the learning model, and further inputting the ground truth label of the upwelling region to the learning model; and a detection unit 17 for acquiring a classification result of an upwelling region in the ocean from the learning model by inputting sea surface temperature data and chlorophyll-a concentration data of the ocean acquired from a satellite to the learning model.

Description

Detection device and detection method

[0001] The present disclosure relates to a detection device and a detection method.

[0002] There is technology to detect upwelling areas. Upwelling is a phenomenon in which deep ocean water rises to the surface, and upwelling areas become ideal fishing grounds. Technology to detect upwelling areas is extremely important, especially for the purpose of tracking trends in fish catches.

[0003] Conventional detection methods rely on unsupervised learning, which involves rule-based labeling of upwelling regions, followed by supervised learning using a small amount of data, which compares the learning results with manual labeling results (see non-patent document 1).

[0004] Furthermore, previous detection methods relied on unsupervised learning to detect upwelling regions using sea surface temperature (SST) data, chlorophyll-a (Chl-a) concentration data, and their indices.

[0005] Zineb El Abidi and 4 others, “An Efficient Detection of Moroccan CoastalUpwelling Based on Fusion of Chlorophyll-a and Sea Surface Temperature Images With a New Validation Index”, IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 18, NO. 8, August 2021, p.1322-p.1326

[0006] However, conventional detection methods involved labeling upwelling areas based on unknown rules, and did not ensure sufficient expert labeling, resulting in low accuracy in detecting upwelling areas.

[0007] Furthermore, since upwelling regions are only detected using SST data, Chl-a concentration data, and their indices, there was also the problem of low accuracy in detecting upwelling regions.

[0008] The present disclosure has been made in consideration of the above circumstances, and an object of the present disclosure is to provide a technology that can improve the detection accuracy of upwelling regions.

[0009] A detection device according to one embodiment of the present disclosure is a detection device for detecting upwelling regions, and includes: a simulation unit that simulates sea surface temperature data, chlorophyll-a concentration data, and ocean current velocity data for each region contained in the ocean; a preprocessing unit that generates a correct label for the upwelling region based on the vertical velocity contained in the ocean current velocity data; a classification unit that inputs the simulated sea surface temperature data and chlorophyll-a concentration data into a learning model that inputs the sea surface temperature data and chlorophyll-a concentration data and outputs a classification result for the upwelling region, and further inputs the correct label for the upwelling region to train the learning model; and a detection unit that inputs the ocean surface temperature data and chlorophyll-a concentration data acquired from a satellite into the learning model and obtains a classification result for the upwelling region in the ocean from the learning model.

[0010] A detection method according to one embodiment of the present disclosure is a method for detecting upwelling regions, in which a detection device simulates sea surface temperature data, chlorophyll-a concentration data, and ocean current velocity data for each region contained in the ocean, generates a correct label for the upwelling region based on the vertical velocity contained in the ocean current velocity data, inputs the simulated sea surface temperature data and chlorophyll-a concentration data into a learning model that inputs the sea surface temperature data and chlorophyll-a concentration data and outputs a classification result for the upwelling region, further inputs the correct label for the upwelling region, and trains the model, inputs the sea surface temperature data and chlorophyll-a concentration data of the ocean acquired from a satellite into the learning model, and obtains a classification result for the upwelling region in the ocean from the learning model.

[0011] According to the present disclosure, a technique can be provided that can improve the detection accuracy of upwelling regions.

[0012] Fig. 1 is a diagram showing a functional block configuration of a detection device. Fig. 2 is a diagram showing a processing flow of the detection device (during learning). Fig. 3 is a diagram showing a processing flow of the detection device (during detection). Fig. 4 is a diagram showing a hardware configuration of the detection device.

[0013] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the description of the drawings, the same parts are designated by the same reference numerals and the description thereof will be omitted.

[0014] [Summary] The present disclosure relates to the technical fields of oceanography and fisheries science, and in particular to a technique for detecting upwelling regions using satellite data and current simulations.

[0015] This paper presents a technique for detecting upwelling regions that combines machine learning and satellite data. By comprehensively analyzing multiple data sources, new criteria for detecting upwelling regions are realized in a learning model.

[0016] Specifically, a learning model is used that inputs SST data and Chl-a concentration data and outputs classification results for upwelling regions. Ocean current velocity data is also simulated, and correct labels for upwelling regions are generated based on the vertical velocity contained in the ocean current velocity data. These correct labels for upwelling regions are then input into the learning model for training. In this way, correct labels for upwelling regions based on vertical velocity are input into the learning model for training, resulting in a learning model that can classify upwelling regions with high accuracy.

[0017] In addition, satellite data (SST data and Chl-a concentration data) obtained from satellites are input into the learning model to classify (detect) upwelling regions. In this way, upwelling regions are classified using actual satellite data measured over a wide area and comprehensively by satellites in the sky, allowing for more accurate classification of upwelling regions.

[0018] These findings provide a technology that can improve the accuracy of detecting upwelling regions, and are expected to contribute to the protection of marine ecosystems and the sustainable management of fisheries resources.

[0019] [Functions of the Detection Apparatus] FIG. 1 is a diagram showing the functional block configuration of the detection apparatus 1. As shown in FIG.

[0020] The detection device 1 includes a simulation unit 11 , a preprocessing unit 12 , an integration unit 13 , a classification unit 14 , a selection unit 15 , an acquisition unit 16 , a detection unit 17 , and a storage unit 18 .

[0021] The simulation unit 11 has the function of simulating the SST data, Chl-a concentration data, and ocean current velocity data for each region included in the target ocean, based on the initial values ​​of the SST data, the initial values ​​of the Chl-a concentration data, the seafloor topography data, the initial values ​​of the underwater temperature data, and the initial values ​​of the salinity concentration data.

[0022] The region is not limited to a region having an area, but also includes a point that is considered to have no area. For example, it is a region that includes a grid when the target ocean is divided into a grid in a simulation, a grid, and the area including the periphery of the grid.

[0023] The preprocessing unit 12 has a function of preprocessing the simulated SST data and Chl-a concentration data, respectively, such as unifying the spatial resolution and normalizing the data values.

[0024] The preprocessing unit 12 has a function of generating a correct label for the upwelling region based on the vertical velocity included in the simulated ocean current velocity data.

[0025] The integration unit 13 has a function of integrating the pre-processed SST data and Chl-a concentration data for each region.

[0026] The classification unit 14 has a function of inputting the integrated data of SST data and Chl-a concentration data into a plurality of different learning models that input the integrated data of SST data and Chl-a concentration data and output classification results of upwelling regions, respectively, and further inputting the correct labels of the above-mentioned upwelling regions as training data into each of the learning models, and causing them to learn.

[0027] The selection unit 15 has a function of selecting, from among a plurality of different learning models, the learning model with the highest classification accuracy for upwelling regions as the trained model.

[0028] The acquisition unit 16 has a function of acquiring SST data and Chl-a concentration data of the target ocean from a satellite.

[0029] The detection unit 17 has the function of inputting SST data and Chl-a concentration data of the target ocean obtained from a satellite into a trained model and obtaining classification results of upwelling regions in the target ocean from the trained model.

[0030] The memory unit 18 has a function of storing data necessary for detecting an upwelling region in the target ocean. For example, the memory unit 18 stores initial values ​​of SST data, initial values ​​of Chl-a concentration data, seafloor topography data, initial values ​​of underwater temperature data, and initial values ​​of salinity concentration data for the target ocean. These data values ​​are associated with location information of the region. The memory unit 18 may be a memory unit included in the detection device 1, or may be a server or database on the Internet, such as a server of the Japan Meteorological Agency.

[0031] [Operation of the Detection Apparatus (During Learning)] FIG. 2 is a diagram showing a processing flow of the detection apparatus 1 (during learning).

[0032] The storage unit 18 stores in advance the initial values ​​of SST data, Chl-a concentration data, bathymetry data, underwater temperature data, and salinity data for the target ocean. These data can be acquired, for example, from a server on the Internet or a satellite.

[0033] Step S101: The simulation unit 11 reads the initial values ​​of the SST data, Chl-a concentration data, bathymetry data, underwater temperature data, and salinity data of the target ocean from the storage unit 18, and simulates the SST data, Chl-a concentration data, and ocean current velocity data (horizontal velocity u·v, vertical velocity w) of each grid in the target ocean using a predetermined simulator. The predetermined simulator is, for example, the Massachusetts Institute of Technology General Circulation Model (MITGCM).

[0034] Step S102: The preprocessing unit 12 preprocesses the simulated SST data and Chl-a concentration data, respectively, by, for example, standardizing the spatial resolution and normalizing the data values.

[0035] Step S103: The preprocessing unit 12 generates a correct label for the upwelling region based on the vertical velocity w included in the simulated ocean current velocity data. For example, the preprocessing unit 12 determines for each grid whether the vertical velocity w of the ocean current velocity data is equal to or greater than a vertical velocity threshold indicating upwelling, and associates a correct label for the upwelling region with a grid having a vertical velocity w equal to or greater than the vertical velocity threshold.

[0036] Step S104: The integration unit 13 integrates the pre-processed SST data and Chl-a concentration data with each other for each grid.

[0037] Step S105: The detection device 1 prepares in advance a plurality of different upwelling region classification learning models for classifying upwelling regions. The upwelling region classification learning model is, for example, a convolutional neural network (CNN), which is a learning model that inputs integrated data of SST data and Chl-a concentration data and outputs a classification result of upwelling regions. The classification result of upwelling regions is, for example, a map of the ocean to be detected, an upwelling grid (upwelling region), a grid flow velocity (flow velocity of the region), and a grid upwelling probability (upwelling probability of the region).

[0038] The classification unit 14 inputs the integrated data of the SST data and Chl-a concentration data integrated in step S104 into each of the multiple upwelling region classification learning models. Next, the classification unit 14 updates the model parameters of each learning model based on the upwelling region classification results output from each of the multiple upwelling region classification learning models. Next, the classification unit 14 causes a learning device to learn the correct upwelling region in the detection target ocean using the correct upwelling region label generated in step S103, and inputs the correct upwelling region into each of the multiple upwelling region classification learning models. Thereafter, each of the multiple upwelling region classification learning models correctly learns the upwelling region in the detection target ocean based on the input correct upwelling region.

[0039] The classification unit 14 repeats this series of processes to recursively train multiple upwelling region classification learning models.

[0040] Step S106: The selection unit 15 evaluates the classification accuracy of the upwelling region based on the classification results of the upwelling region output from each of the multiple upwelling region classification learning models, and selects the upwelling region classification learning model with the highest classification accuracy of the upwelling region as the trained model.

[0041] [Operation of the Detection Apparatus (During Detection)] FIG. 2 is a diagram showing a processing flow of the detection apparatus 1 (during detection).

[0042] Step S201: The acquisition unit 16 acquires SST data and Chl-a concentration data of the target ocean from a satellite.

[0043] Step S202: The pre-processing unit 12 pre-processes the SST data and Chl-a concentration data of the target ocean acquired from the satellite.

[0044] Step S203: The integration unit 13 integrates the pre-processed SST data and Chl-a concentration data with each other for each grid.

[0045] Step S204: The detection unit 17 inputs the integrated data of the integrated SST data and Chl-a concentration data into the trained model of step S106 (the upwelling region classification trained model with the highest upwelling region classification accuracy), and obtains a classification result of the upwelling region in the detection target ocean from the trained model. The detection unit 17 outputs this upwelling region classification result as the classification result of the upwelling region in the detection target ocean.

[0046] This completes the detection process. Satellite data from the past five years in a specific ocean area is collected, and the actual SST data and Chl-a concentration data are used as input data for the trained model. The upwelling areas and upwelling area probabilities for the past five years are mapped on a map and output in chronological order. Users can use this map to analyze seasonal and yearly fluctuations in upwelling areas, and use this information to help manage fisheries resources.

[0047] [Effects] According to this embodiment, SST data, Chl-a concentration data, and ocean current velocity data for each region included in the ocean are simulated, and correct labels for upwelling regions are generated based on the vertical velocity included in the ocean current velocity data. The simulated SST data and Chl-a concentration data are input into an upwelling region classification learning model that inputs the SST data and Chl-a concentration data and outputs classification results for upwelling regions, and the correct labels for upwelling regions are further input and trained. Ocean SST data and Chl-a concentration data acquired from a satellite are input into the upwelling region classification learning model, and classification results for upwelling regions in the ocean are obtained from the upwelling region classification learning model, thereby providing a technology that can improve the accuracy of detecting upwelling regions.

[0048] [Others] The present disclosure is not limited to the above-described embodiments, and various modifications are possible within the scope of the gist of the present disclosure.

[0049] The detection device 1 of the present embodiment described above can be realized, for example, by using a general-purpose computer system including a CPU 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906, as shown in Fig. 4. The memory 902 and the storage 903 are storage devices. In the computer system, the CPU 901 executes a predetermined program loaded on the memory 902, thereby realizing each function of the detection device 1.

[0050] The detection device 1 may be implemented by one computer, or by multiple computers, or may be a virtual machine implemented on a computer.

[0051] The program for the detection device 1 can be stored in a computer-readable recording medium such as a HDD, SSD, USB memory, CD, or DVD. The computer-readable recording medium is, for example, a non-transitory recording medium. The program for the detection device 1 can also be distributed via a communication network.

[0052] REFERENCE SIGNS LIST 1 detection device 11 simulation unit 12 preprocessing unit 13 integration unit 14 classification unit 15 selection unit 16 acquisition unit 17 detection unit 18 storage unit 901 CPU 902 memory 903 storage 904 communication device 905 input device 906 output device

Claims

1. A detection device for detecting upwelling regions, comprising: a simulation unit that simulates sea surface temperature data, chlorophyll-a concentration data, and ocean current velocity data for each region contained in the ocean; a preprocessing unit that generates correct labels for upwelling regions based on the vertical velocity contained in the ocean current velocity data; a classification unit that inputs the simulated sea surface temperature data and chlorophyll-a concentration data into a learning model that inputs the sea surface temperature data and chlorophyll-a concentration data and outputs classification results for upwelling regions, and further inputs the correct labels for the upwelling regions into the learning model, and performs learning; and a detection unit that inputs the sea surface temperature data and chlorophyll-a concentration data of the ocean obtained from a satellite into the learning model, and obtains classification results for upwelling regions in the ocean from the learning model.

2. The detection device according to claim 1, wherein the detection unit obtains the classification result of the upwelling region in the ocean from the learning model with the highest classification accuracy of the upwelling region among a plurality of different learning models.

3. The detection device of claim 1, wherein the simulation unit simulates the sea surface temperature data, the chlorophyll-a concentration data, and the ocean current velocity data from initial values ​​of sea surface temperature data, initial values ​​of chlorophyll-a concentration data, seabed topography data, initial values ​​of underwater water temperature data, and initial values ​​of salinity concentration data.

4. A detection method for detecting upwelling regions, comprising: a detection device that simulates sea surface temperature data, chlorophyll-a concentration data, and ocean current velocity data for each region contained in the ocean; generates a correct label for the upwelling region based on the vertical velocity contained in the ocean current velocity data; inputs the simulated sea surface temperature data and chlorophyll-a concentration data into a learning model that inputs the sea surface temperature data and chlorophyll-a concentration data and outputs a classification result for the upwelling region, and further inputs the correct label for the upwelling region to train the model; inputs the sea surface temperature data and chlorophyll-a concentration data for the ocean acquired from a satellite into the learning model, and obtains a classification result for the upwelling region in the ocean from the learning model.

Images