Fish school distribution forecasting method and device based on deep learning and pattern cooperation

Abstract

The invention relates to a fish school distribution forecasting method and device based on deep learning and pattern collaboration, and the method comprises the steps: carrying out the space-time reconstruction of marine environment parameter remote sensing data of a target region, and generating continuous remote sensing data; performing three-dimensional processing on the continuous remote sensing data, and generating a vertical layered marine environment element data set in combination with reanalysis data; a CNN model is used for fitting the relation between the marine environment elements and the fishing time of the fishing boat, and a fish school distribution prediction model is constructed; obtaining the predicted fishing time of the fishing boat; and constructing a fish school index according to the predicted fishing time of the fishing boat. The method has the advantages that the LSTM technology, the hydrodynamic-ecological coupling mode and the CNN model are combined, and multi-source satellite remote sensing data, reanalysis data and fishery data are fully used. By utilizing the method, the fish school distribution forecasting accuracy can be improved, the spatial-temporal dynamic change of fish school distribution can be obtained, and a scientific decision-making tool can be provided for management and protection of fishery resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring of marine environments, and in particular relates to a method and device for predicting fish distribution based on deep learning and pattern collaboration. Background Art

[0002] Predicting the timing, location, and size of fish distribution provides information support for marine fishery planning and is of great significance to coastal management and aquaculture. Due to the unique global status of fisheries, fish forecasting has always been a key research topic. It is not only a technical tool for maintaining global food security and ecological balance, but also a key support for addressing climate change and achieving "blue growth."

[0003] Currently, satellite remote sensing data has become the core data support for large-scale fish distribution forecasts. It inverts key environmental parameters such as sea surface temperature and chlorophyll concentration, combines them with historical fish distribution data in the target area, and constructs predictive models using machine learning algorithms. However, due to technical limitations and observation mechanisms, practical applications still face four difficulties: 1. The spatiotemporal discontinuity of remote sensing data. Due to sensor failure, cloud cover, or data transmission interruptions, remote sensing data is prone to incomplete data coverage, insufficient resolution, and uneven temporal distribution, which reduces the reliability of remote sensing data. 2. The lack of depth information for ocean elements. Existing remote sensing data is presented in two-dimensional form, mainly reflecting the physical or biochemical characteristics of the ocean surface. It fails to cover the vertical dimension of the ocean, making it difficult to fully and accurately reveal the complex three-dimensional response relationships between fish and environmental factors, thus limiting the accuracy and reliability of fish distribution forecasts. 3. The lack of historical fish distribution information. Monitoring technologies such as acoustic monitoring are significantly affected by environmental noise, resulting in low data reliability. Inadequate data integration and sharing also exacerbate the lack of true catch data. ④ The forecast accuracy of traditional numerical models is limited. Fisheries are highly dynamic, but traditional numerical models struggle to capture the dynamics of the marine environment, resulting in low forecast accuracy. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for predicting fish distribution based on deep learning and model collaboration.

[0005] First, a fish distribution prediction method based on deep learning and model collaboration is provided, including:

[0006] Step 1: Reconstruct the remote sensing data of marine environmental parameters in the target area in time and space to generate continuous remote sensing data;

[0007] Step 2: 3D-process the continuous remote sensing data using the ROMS model, and generate a vertically layered marine environmental element dataset by combining it with the reanalysis data;

[0008] Step 3: Use fishing time to represent fish distribution information, use the CNN model to fit the relationship between marine environmental factors and fishing time, and build a fish distribution prediction model;

[0009] Step 4: Input the new marine environmental factors of the target area into the fish distribution prediction model to obtain the predicted fishing time of the fishing boat;

[0010] Step 5: Construct a fish school index based on the predicted fishing time of the fishing boat.

[0011] Preferably, in step 1, the ocean environment parameter remote sensing data includes chlorophyll concentration data, sea surface temperature data and ocean surface salinity data.

[0012] Preferably, in step 1, the interpolation technology and the LSTM model are combined to perform spatiotemporal reconstruction of the remote sensing data of marine environmental parameters.

[0013] Preferably, in step 2, the calculation formula of the ROMS mode includes:

[0014] Continuity equation:

[0015]

[0016] Convection-diffusion equation:

[0017]

[0018] Water state equation:

[0019] ρ=f(C,p)

[0020] Among them, H z refers to the height of each grid; u, v, and Ω refer to the flow velocities in the horizontal and vertical directions, respectively; p is the pressure; ρ and ρ0 are the actual seawater density and the reference density, respectively; ζ is the height of the free sea surface; the overline (-) indicates the averaging time, and the superscript symbol (′) indicates the turbulent disturbance; v and v θ Represent the water viscosity coefficient and diffusion coefficient respectively; D c represents the horizontal diffusion term; C source represents the source and sink terms; f represents the Coriolis force parameter, and g represents the acceleration due to gravity.

[0021] Preferably, in step 5, the calculation formula of the fish school index is:

[0022]

[0023] Among them, image data Refers to the original data value, min(image data ) refers to the minimum value of the original data, max(image date ) refers to the maximum value of the original data.

[0024] In a second aspect, a fish distribution prediction device based on deep learning and pattern collaboration is provided, which is used to execute any of the methods described in the first aspect, including:

[0025] The reconstruction module is used to perform spatiotemporal reconstruction of the remote sensing data of the marine environmental parameters of the target area to generate continuous remote sensing data;

[0026] A three-dimensionalization module is used to process the continuous remote sensing data into three dimensions through the ROMS model and generate a vertically layered marine environmental element dataset in combination with reanalysis data;

[0027] The first construction module is used to characterize fish distribution information based on the fishing time of fishing vessels, and uses the CNN model to fit the relationship between marine environmental factors and the fishing time of fishing vessels to build a fish distribution prediction model;

[0028] an acquisition module, configured to input new marine environmental elements of the target area into the fish distribution prediction model to obtain a predicted fishing time of the fishing vessel;

[0029] The second building module is used to build a fish school index based on the predicted fishing time of the fishing boat.

[0030] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any one of the methods described in the first aspect.

[0031] In a fourth aspect, an electronic device is provided, including:

[0032] Memory, used to store computer programs;

[0033] A processor is used to execute the computer program to implement any method as described in the first aspect.

[0034] The beneficial effects of the present invention are as follows: the present invention first reconstructs the missing pixels in the high-frequency remote sensing time series data through interpolation technology and LSTM model, and then establishes ROMS model to output a multi-level marine environmental element data set, which makes up for the deficiency of remote sensing data in expressing at the depth level, and conducts in-depth analysis and nonlinear fitting of various marine environmental elements and fishing vessel spatiotemporal distribution data through CNN model, and finally constructs fish index to achieve accurate prediction of the spatial position distribution of fish. The method proposed in the present invention combines LSTM technology, hydrodynamic-ecological coupling model and CNN model, and makes full use of multi-source satellite remote sensing data, reanalysis data and fishery data. This method can improve the accuracy of fish distribution forecast, obtain the spatiotemporal dynamic changes of fish distribution, provide scientific decision-making tools for the management and protection of fishery resources, and has important production application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a fish distribution prediction method based on deep learning and model collaboration provided by the present invention;

[0036] Figure 2 The present invention provides a fishing time measurement chart and a prediction chart of fishing vessels; (a) is the measurement chart, and (b) is the prediction chart;

[0037] Figure 3 The real and predicted graphs of the fish school index provided by the present invention are shown in Figure 1; (a) is the real graph, and (b) is the predicted graph;

[0038] Figure 4 This is a schematic diagram of a fish distribution prediction device based on deep learning and model collaboration provided by the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to the following examples. The following examples are provided only to facilitate understanding of the present invention. It should be noted that, without departing from the principles of the present invention, it is possible for a person skilled in the art to make various modifications to the present invention, and such improvements and modifications fall within the scope of the claims of the present invention.

[0040] Example 1:

[0041] To address the spatiotemporal discontinuity of remote sensing data, the lack of layers of marine environmental elements, and the numerous limitations and low accuracy of traditional numerical model forecasts in traditional fishery forecasting methods, Example 1 of this application provides a fish distribution forecasting method based on deep learning and model collaboration, using multiple public satellite remote sensing data, reanalysis data, and fishery data as data sources, and implemented on the Python 3.8 platform. The study area is located in the East China Sea, with a latitude and longitude range of 26°N-34°N and 119°E-129°E.

[0042] Specifically, such as Figure 1 As shown, the method provided by this application includes:

[0043] Step 1: Reconstruct the remote sensing data of marine environmental parameters in the target area in time and space to generate continuous remote sensing data.

[0044] Specifically, the L3 chlorophyll concentration, sea surface temperature of the MODIS satellite and the ocean surface salinity data products of the SMOS satellite in the study area were screened, and the remote sensing time domain was reconstructed through interpolation and LSTM model to obtain a total of 122 product data images from August to September 2022 and August to September 2023.

[0045] In addition, in step 1, the interpolation technology is combined with the LSTM (long short-term memory network) model to perform spatiotemporal reconstruction of the remote sensing data of marine environmental parameters.

[0046] The calculation formula of the interpolation method is:

[0047]

[0048] Where: x(t i ) is the time point t i The interpolation result of x(t i -1) and x(t i +1) is the time point t i -1 and t i +1 known data value.

[0049] The calculation formula of the LSTM model is:

[0050]

[0051] j=tanh(W j [h t-1 ,x t ]+b j )

[0052]

[0053] Step 2: Process the continuous remote sensing data into three dimensions using the ROMS model, and generate a vertically layered marine environmental element dataset by combining it with the reanalysis data.

[0054] In step 2, based on the established ROMS model, three-dimensional remote sensing environmental element data, and collaborative reanalysis data, vertical ocean environmental elements were constructed and optimized to construct a three-dimensional ocean environmental element dataset.

[0055] Specifically, observation data and reanalysis information from August to September 2022 will be obtained, the ROMS model will be established, and remote sensing data will be processed in three dimensions to generate a vertically layered marine environmental element data set including ocean temperature, salinity, ocean currents, chlorophyll a and primary productivity.

[0056] In step 2, the calculation formula of the ROMS mode includes:

[0057] Continuity equation:

[0058]

[0059] Convection-diffusion equation:

[0060]

[0061] Water state equation:

[0062] ρ=f(C,p)

[0063] Among them, H Z refers to the height of each grid; u, v, and Ω refer to the flow velocities in the horizontal and vertical directions, respectively; p is the pressure; ρ and ρ0 are the actual seawater density and the reference density, respectively; ζ is the height of the free sea surface; the overline (-) indicates the averaging time, and the superscript symbol (′) indicates the turbulent disturbance; v and v θ Represent the water viscosity coefficient and diffusion coefficient respectively; D c represents the horizontal diffusion term; C source represents the source and sink terms; f represents the Coriolis force parameter, and g represents the acceleration due to gravity.

[0064] Step 3: Use the fishing time of fishing boats to represent the fish distribution information, use the CNN model to fit the relationship between marine environmental factors and the fishing time of fishing boats, and build a fish distribution prediction model.

[0065] Step 4: Input the new marine environmental elements of the target area into the fish distribution prediction model to obtain the predicted fishing time of the fishing boat.

[0066] Step 5: Construct a fish school index based on the predicted fishing time of the fishing boat.

[0067] Example 2:

[0068] Based on Example 1, Example 2 of the present application provides a more specific fish distribution prediction method based on deep learning and model collaboration, including:

[0069] Step 1: Reconstruct the remote sensing data of marine environmental parameters in the target area in time and space to generate continuous remote sensing data.

[0070] Step 2: Three-dimensional processing of the continuous remote sensing data by the ROMS model, combined with reanalysis data to generate a vertical layered marine environmental element dataset.

[0071] Step 3: Representing fish distribution information with fishing duration, using a CNN model to fit the relationship between marine environmental elements and fishing duration, and constructing a fish distribution prediction model.

[0072] In Step 3, fishing duration is used to represent fish distribution information, solving the problem of missing fish measurement data. The marine environmental element dataset and fishing duration data are trained and fitted using a CNN model to obtain a fish distribution prediction model.

[0073] Specifically, fishing duration data for the study area is obtained from the Global Fishing Watch platform, and the marine environmental element dataset for August-September 2022 is used as the independent variable, and the fishing duration data for August-September 2022 is used as the dependent variable to train the CNN model and construct an optimal fish distribution prediction model.

[0074] In Step 3, the CNN model includes the following formula:

[0075] The calculation formula for the convolution operation is:

[0076]

[0077] Where X is the input feature map, W is the convolution kernel weight, b is the bias term, and Y is the output feature map.

[0078] Downsampling operation, the maximum pooling calculation formula is:

[0079] Y i,j,k =max(X 2i,2j,k ,X 2i+1,2j,k ,X 2i,2j+1,k ,X 2i+1,2j+1,k )

[0080] Where X is the input feature map and Y is the pooled feature map.

[0081] Map the feature maps extracted by the convolution layer and the pooling layer to the target output space, and the calculation formula of the full connection layer is:

[0082] y=f(Wx+b)

[0083] Where X is the input vector, W is the weight matrix, b is the bias, and f is the activation function (such as ReLU or Softmax).

[0084] Step 4: Input the new marine environmental elements of the target area into the fish distribution prediction model to obtain the predicted fishing duration.

[0085] Specifically, such as Figure 2 As shown in the figure, the marine environmental factor data from August to September 2023 were input into the fish distribution prediction model to obtain the prediction results of fishing time of fishing vessels from August to September 2023, and the accuracy was evaluated.

[0086] Step 5: Construct a fish school index based on the predicted fishing time of the fishing boat.

[0087] Specifically, such as Figure 3 As shown, this application constructs a fish index based on deep learning and model collaboration. In the process of index construction, remote sensing data, numerical patterns, LSTM and CNN models are comprehensively utilized to achieve accurate prediction of fish distribution under the condition of missing true values, and finally obtain the distribution range of fish.

[0088] In step 5, the calculation formula of the fish school index is:

[0089]

[0090] Among them, image data Refers to the original data value, min(image data ) refers to the minimum value of the original data, max(image data ) refers to the maximum value of the original data.

[0091] For example, Table 1 is a daily fish index accuracy evaluation table from August 1 to September 30, 2023.

[0092] Table 1

[0093]

[0094]

[0095] It should be noted that the parts in this embodiment that are the same or similar to those in Example 1 can be referenced to each other and will not be described in detail in this application.

[0096] Example 3:

[0097] Based on Example 2, Example 3 of the present application provides a fish distribution prediction device based on deep learning and model collaboration, including:

[0098] The reconstruction module is used to perform spatiotemporal reconstruction of the remote sensing data of the marine environmental parameters of the target area to generate continuous remote sensing data;

[0099] A three-dimensionalization module is used to process the continuous remote sensing data into three dimensions through the ROMS model and generate a vertically layered marine environmental element dataset in combination with reanalysis data;

[0100] The first construction module is used to characterize fish distribution information based on the fishing time of fishing vessels, and uses the CNN model to fit the relationship between marine environmental factors and the fishing time of fishing vessels to build a fish distribution prediction model;

[0101] an acquisition module, configured to input new marine environmental elements of the target area into the fish distribution prediction model to obtain a predicted fishing time of the fishing vessel;

[0102] The second building module is used to build a fish school index based on the predicted fishing time of the fishing boat.

[0103] It should be noted that the system provided in this embodiment is a device corresponding to the method provided in Example 2. Therefore, the parts in this embodiment that are the same or similar to those in Example 2 can be referenced to each other and will not be repeated in this application.

[0104] In summary, this application generates a vertically layered multi-ocean environmental element dataset through high-quality remote sensing data reconstruction and ROMS model, characterizes fish distribution information by fishing duration of fishing vessels, and uses CNN model to establish a fish distribution prediction model, thereby constructing a fish distribution index, realizing accurate prediction of fish distribution, and providing a scientific basis for the management and protection of fishery resources.

Claims

1. A fish distribution prediction method based on deep learning and model collaboration, characterized in that: include: Step 1: Reconstruct the remote sensing data of marine environmental parameters in the target area in time and space to generate continuous remote sensing data; Step 2: 3D-process the continuous remote sensing data using the ROMS model, and generate a vertically layered marine environmental element dataset by combining it with reanalysis data; Step 3: Use fishing time to represent fish distribution information, use the CNN model to fit the relationship between marine environmental factors and fishing time, and build a fish distribution prediction model; Step 4: Input the new marine environmental factors of the target area into the fish distribution prediction model to obtain the predicted fishing time of the fishing boat; Step 5: Construct a fish school index based on the predicted fishing time of the fishing boat.

2. The fish distribution prediction method based on deep learning and pattern collaboration according to claim 1 is characterized in that: In step 1, the ocean environment parameter remote sensing data includes chlorophyll concentration data, sea surface temperature data and ocean surface salinity data.

3. The fish distribution prediction method based on deep learning and pattern collaboration according to claim 2 is characterized in that: In step 1, the interpolation technology and LSTM model are combined to reconstruct the spatiotemporal remote sensing data of marine environmental parameters.

4. The fish distribution prediction method based on deep learning and pattern collaboration according to claim 3 is characterized in that: In step 2, the calculation formula of the ROMS mode includes: Continuity equation: Convection-diffusion equation: Water state equation: ρ=f(C,p) Among them, H z refers to the height of each grid; u, v, and Ω refer to the flow velocities in the horizontal and vertical directions, respectively; p is the pressure; ρ and ρ0 are the actual seawater density and the reference density, respectively; ζ is the height of the free sea surface; the overline (-) indicates the averaging time, and the superscript symbol (′) indicates the turbulent disturbance; v and v θ Represent the water viscosity coefficient and diffusion coefficient respectively; D c represents the horizontal diffusion term; C source represents the source and sink terms; f represents the Coriolis force parameter, and g represents the acceleration due to gravity.

5. The fish distribution prediction method based on deep learning and pattern collaboration according to claim 4 is characterized in that: In step 5, the calculation formula of the fish school index is: Among them, image data Refers to the original data value, min(image data ) refers to the minimum value of the original data, max(image data ) refers to the maximum value of the original data.

6. A fish distribution prediction device based on deep learning and model collaboration, characterized in that: Used to perform the method according to any one of claims 1 to 5, comprising: The reconstruction module is used to perform spatiotemporal reconstruction of the remote sensing data of the marine environmental parameters of the target area to generate continuous remote sensing data; A three-dimensionalization module is used to process the continuous remote sensing data into three dimensions through the ROMS model and generate a vertically layered marine environmental element dataset in combination with reanalysis data; The first construction module is used to characterize fish distribution information based on the fishing time of fishing vessels, and uses the CNN model to fit the relationship between marine environmental factors and the fishing time of fishing vessels to build a fish distribution prediction model; an acquisition module, configured to input new marine environmental elements of the target area into the fish distribution prediction model to obtain a predicted fishing time of the fishing vessel; The second building module is used to build a fish school index based on the predicted fishing time of the fishing boat.

7. A computer storage medium, characterized in that The computer storage medium stores a computer program; when the computer program is run on a computer, the computer executes the method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 5.

Images