Fishery resource calculation method and system for marine ranch

Abstract

The invention discloses a fishery resource calculation method and system for a marine ranch, and the method not only carries out the fish school recognition and statistics of a sonar image through an image recognition technology, but also predicts the number of planktons in water through water environment data, thereby predicting the number of fish schools. By utilizing the characteristic that the plankton quantity is associated with the fish school quantity, finally, the fish school quantity obtained by image recognition and the fish school quantity obtained by prediction based on the plankton quantity are comprehensively analyzed and calculated to obtain the final fish school quantity, so that the fish school quantity statistical accuracy of the marine ranch is improved, and the fish school quantity statistical efficiency is improved. The problem that in the prior art, the accuracy of the fish school number obtained through final statistics is low due to low fish school image recognition precision is solved.

Description

Technical Field

[0001] The present invention relates to the field of fishery resource calculation, and in particular to a method and system for calculating fishery resources of a marine ranch. Background Art

[0002] Due to overfishing and extensive aquaculture in marine fisheries, the ecological environment of some sea areas has been damaged and fishery resources have declined. In addition, with the increasing demand for aquatic products and the increasing concern for the sustainable development of marine fisheries, it is necessary to scientifically evaluate and manage the fishery resources of marine ranches to ensure the stability and continuous growth of aquatic resources while protecting the marine ecosystem. Therefore, it is necessary to accurately count the number of fish in marine ranches.

[0003] The existing technology usually only uses image recognition algorithms to identify the underwater fish sonar images collected, and then counts the number of fish in the marine ranch. However, due to the limitations of the acoustic imaging mechanism and noise pollution, sonar images generally have problems such as low contrast and blurred edges, which easily lead to poor quality sonar images. Therefore, the existing technology will have low accuracy in the final number of fish counts due to the low accuracy of fish image recognition. Summary of the invention

[0004] The present invention provides a method and system for calculating fishery resources in an ocean ranch, which can solve the problem in the prior art that the accuracy of the number of fish schools finally counted is low due to the low accuracy of fish school image recognition.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for calculating fishery resources of a marine ranch, comprising: Divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; Inputting the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish school identification in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; Inputting the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; Based on the number of fish schools identified and predicted in each ocean layer, the comprehensive number of fish schools in each ocean layer is calculated, and based on the comprehensive number of fish schools in each ocean layer, the total number of fish schools to be counted in the marine ranch is calculated.

[0006] Furthermore, the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer to obtain the number of fish school identifications in each ocean layer, including: The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer to obtain the local feature map of each ocean layer; The global feature maps of each ocean layer are extracted by using the multi-head autonomous intention layer and feedforward neural network layer of the Transformer encoder in the global feature extraction module. The local feature map and the global feature map of each ocean layer are fused respectively through the feature fusion module to obtain the fused feature map of each ocean layer; The backbone network layer, neck network layer and detection head layer of the YOLOv5 recognition module are used to identify and count the fused feature maps of each ocean layer to obtain the number of fish schools identified in each ocean layer.

[0007] Furthermore, the trained fish population prediction model first predicts the number of plankton in each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the number of fish according to the predicted number of plankton, including: The plankton quantity is predicted according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted plankton quantity of each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; According to the predicted number of plankton in each ocean layer and the real-time water environment data, the fish population is predicted through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

[0008] Furthermore, the plankton quantity prediction is performed according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted number of plankton in each ocean layer, including: The real-time water environment data of each ocean layer is preprocessed through the input layer of the plankton prediction module to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; According to the real-time water environment data processed by each ocean layer, the final hidden state of the LSTM unit is calculated and output through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit in the first hidden layer of the plankton prediction module; The hidden state of the LSTM unit is set as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit of the first hidden layer in the plankton prediction module; According to the final hidden state of the GRU unit, the plankton quantity is predicted through the fully connected unit of the output layer in the plankton prediction module to obtain the predicted plankton quantity in each ocean layer.

[0009] Furthermore, the calculation formula for the comprehensive number of fish schools in each ocean layer is: ; in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks for the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ; in, The total number of marine ranch fish stocks to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

[0010] Based on the above method embodiment, the present invention provides a corresponding system embodiment; An embodiment of the present invention provides a fishery resource calculation system for a marine ranch, comprising: a data acquisition module, a fish population identification module, a fish population prediction module, and a fish population total population calculation module; The data acquisition module is used to divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; The fish school number recognition module is used to input the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish schools recognized in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The fish population prediction module is used to input the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, so as to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; The total fish number calculation module is used to calculate the comprehensive number of fish schools in each ocean layer according to the number of fish school identification and the predicted number of fish schools in each ocean layer, and calculate the total number of fish schools to be counted in the marine ranch according to the comprehensive number of fish schools in each ocean layer.

[0011] Furthermore, the fish school image recognition model includes: a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer of the local feature extraction module to obtain a local feature map of each ocean layer; The global feature extraction module extracts global features of the image from the local feature maps of each ocean layer through the multi-head autonomous intention layer and the feedforward neural network layer of the Transformer encoder in the global feature extraction module to obtain the global feature map of each ocean layer; The feature fusion module fuses the local feature map and the global feature map of each ocean layer respectively through the feature fusion module to obtain a fused feature map of each ocean layer; The YOLOv5 recognition module performs fish identification and quantity statistics on the fusion feature graph of each ocean layer through the backbone network layer, the neck network layer and the detection head layer of the YOLOv5 recognition module to obtain the number of fish schools identified in each ocean layer.

[0012] Furthermore, the fish population prediction model includes: a plankton prediction module and a fish population prediction module; The plankton prediction module is used to predict the number of plankton according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module, so as to obtain the predicted number of plankton in each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; The fish population prediction module is used to predict the number of fish populations in each ocean layer based on the predicted number of plankton in each ocean layer and real-time water environment data, through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

[0013] Furthermore, the plankton prediction module includes an input layer, a first hidden layer and an output layer; the first hidden layer includes an LSTM unit and a GRU unit; The input layer of the plankton prediction module is used to perform data preprocessing on the real-time water environment data of each ocean layer to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; The LSTM unit of the first hidden layer in the plankton prediction module is used to calculate and output the final hidden state of the LSTM unit through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit of the first hidden layer in the plankton prediction module according to the real-time water environment data processed by each ocean layer; The GRU unit of the first hidden layer in the plankton prediction module is used to set the hidden state of the LSTM unit as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit in the first hidden layer of the plankton prediction module; The output layer in the plankton prediction module is used to predict the number of plankton according to the final hidden state of the GRU unit through the fully connected units of the output layer in the plankton prediction module to obtain the predicted number of plankton in each ocean layer.

[0014] Furthermore, the calculation formula for the comprehensive number of fish schools in each ocean layer is: ; in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks in the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ; in, The total number of fish stocks in the marine ranch to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

[0015] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: the technical solution of the present invention first divides the marine ranch into regional depths to obtain several marine layers of different depths; secondly, the constructed fish school image recognition model is used to perform local feature extraction and global feature extraction on the underwater sonar images of each marine layer of the marine ranch to obtain local feature maps and global feature maps of the underwater sonar images of each marine layer, the extracted feature maps are fused, and the image recognition module based on the YOLOv5 recognition algorithm is used to perform fish school recognition and quantity statistics on the fused feature maps to obtain the number of fish school recognition in each marine layer; then, a fish school quantity prediction model including a plankton prediction module and a fish school quantity prediction module is constructed to first predict the number of plankton in each marine layer according to the water environment data of each marine layer, and then The number of fish is predicted based on the predicted number of plankton, and the predicted number of fish in each ocean layer is obtained; finally, the number of fish identified in each ocean layer and the predicted number of fish are comprehensively analyzed and calculated to obtain the final number of fish in the ocean ranch; that is, the present invention not only uses image recognition technology to perform fish identification statistics on sonar images, but also predicts the number of plankton in the water through water environment data, and then predicts the number of fish, and utilizes the characteristic that there is a correlation between the number of plankton and the number of fish. Finally, the number of fish obtained by image recognition and the number of fish predicted based on the number of plankton are comprehensively analyzed and calculated to obtain the final number of fish, thereby improving the accuracy of fish number statistics in the ocean ranch and solving the problem of low accuracy of fish image recognition in the prior art, which leads to low accuracy of the final number of fish obtained by statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1: A flow chart of the steps of a method for calculating fishery resources of a marine ranch provided by an embodiment of the present invention; Figure 2 : A system structure diagram of a fishery resource calculation system for a marine ranch provided by an embodiment of the present invention; Figure 3 : A model structure diagram of a fish school image recognition model provided by an embodiment of the present invention; Figure 4 : A model structure diagram of a fish population prediction model provided by an embodiment of the present invention; DETAILED DESCRIPTION The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] In the description of the present invention, it should be understood that the terms “first” and “second” are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0018] Embodiment 1: Reference Figure 1 , is a flow chart of the steps of a method for calculating fishery resources of a marine ranch provided by an embodiment of the present invention, the method at least comprising the following steps S1 to S4: Step S1: Divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; In this embodiment, according to the water depth of the ocean ranch, several ocean layers with different depth ranges are set, and high-resolution multi-beam sonar equipment and water environment monitoring buoys equipped with multiple sensors are deployed in the center and edge areas of each ocean layer to collect real-time underwater sonar images and real-time water environment data of each ocean layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; the sensors include light intensity sensors, water temperature sensors, salinity sensors and nutrient concentration sensors.

[0019] In this embodiment, after collecting the real-time underwater sonar images and real-time water environment data of each ocean layer, it also includes: performing image preprocessing on the real-time underwater sonar images of each ocean layer; wherein the image preprocessing includes but is not limited to image denoising, image enhancement and image set correction.

[0020] Step S2: inputting the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish school identification in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; In this embodiment, refer to Figure 3 , is a model structure diagram of a fish school image recognition model provided by an embodiment of the present invention, wherein the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar image of each ocean layer to obtain the number of fish school identifications in each ocean layer, including: The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer to obtain the local feature map of each ocean layer; Exemplarily, the local feature extraction module may adopt a CNN algorithm, and the calculation steps of the convolution layer, the pooling layer and the activation function layer of the CNN algorithm are: First, set the underwater sonar image to , the size is (width W, height H and number of channels C). After the first convolution layer, the convolution kernel size is set to , the step size is set to , fill set to , the feature map output by the convolutional layer The size is (in is the number of channels output by the convolutional layer), local feature map The calculation formula is ;in, is the activation function; Then after the pooling layer, the pooling kernel size is set to , the step size is set to , the pooling layer outputs the feature map The size is , the pooling layer outputs the feature map The calculation formula is ; After several convolutional layers, pooling layers, and activation function layers, the local feature map output by CNN is finally obtained. , whose size is .

[0021] The global feature maps of each ocean layer are extracted by using the multi-head autonomous intention layer and feedforward neural network layer of the Transformer encoder in the global feature extraction module. Exemplarily, the calculation steps of the multi-head autonomous intention layer and the feedforward neural network layer of the Transformer encoder in the global feature extraction module are: First, the local feature map Perform dimension transformation and stretch it into a two-dimensional sequence , the shape is (in , ), and add positional coding ,get .

[0022] Then Input into the Transformer encoder, for the multi-head self-attention layer, let the number of heads be , query matrix , key matrix , value matrix Through linear transformation respectively, we get: , and (in , and is the preset weight matrix), the calculation formula of the attention score is (in ), multi-head attention output (in , is the preset output weight matrix); The feature map output by the multi-head autonomous intention layer is fed into the feedforward neural network layer for full connection processing and nonlinear transformation, and finally the global feature map output by the Transformer is obtained.

[0023] The local feature map and the global feature map of each ocean layer are fused respectively through the feature fusion module to obtain the fused feature map of each ocean layer; The backbone network layer, neck network layer and detection head layer of the YOLOv5 recognition module are used to identify and count the fused feature maps of each ocean layer to obtain the number of fish schools identified in each ocean layer.

[0024] In this embodiment, first, feature extraction is performed on the fused feature map of each ocean layer through the backbone network layer of the YOLOv5 recognition module to obtain the feature map of the fused feature map of each ocean layer; then, the feature map of the fused feature map of each ocean layer is fused and adjusted through the neck network layer of the YOLOv5 recognition module to generate a multi-scale feature map of the fused feature map of each ocean layer; finally, fish recognition and quantity statistics are performed on the multi-scale feature map of the fused feature map of each ocean layer through the detection head layer of the YOLOv5 recognition module to obtain the number of fish schools recognized in each ocean layer.

[0025] In this embodiment, the model training steps of the fish school image recognition model are: Obtain historical underwater sonar images tagged with actual numbers of fish schools in each ocean layer of the sea ranch; Performing image preprocessing on historical underwater sonar images with actual number labels of fish schools in each marine layer of the marine ranch; wherein the image preprocessing includes but is not limited to image denoising, image enhancement and image set correction; For each ocean layer, the local feature extraction of the historical underwater sonar images with the actual number of fish labels after image preprocessing is performed through the convolution layer, pooling layer and activation function layer of the local feature extraction module in the fish image recognition model to be trained, and the local feature map of each ocean layer is obtained; The global feature maps of each ocean layer are extracted from the local feature maps of each ocean layer through the multi-head autonomous intention layer and feedforward neural network layer of the Transformer encoder of the global feature extraction module in the fish school image recognition model to be trained, and the global feature maps of each ocean layer are obtained; The local feature map and the global feature map of each ocean layer are fused respectively through the feature fusion module in the fish school image recognition model to be trained to obtain the fused feature map of each ocean layer; The backbone network layer, neck network layer and detection head layer of the YOLOv5 recognition module in the fish school image recognition model to be trained are used to identify and count the number of fish schools in each ocean layer through the fusion feature map of each ocean layer, and the number of fish school identification in each ocean layer is obtained; According to the number of fish schools identified in each ocean layer and the actual number of fish schools, the loss value is calculated through the loss function, and the model parameters of the fish school image recognition model to be trained are optimized according to the loss value until the loss value converges to obtain a trained fish school image recognition model; wherein the loss function includes but is not limited to the mean square error loss function, the mean absolute error loss function and the binary cross entropy loss function; the optimization process includes but is not limited to the stochastic gradient descent optimization algorithm and the Adam optimizer.

[0026] Step S3: inputting the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; In this embodiment, refer to Figure 4 , is a model structure diagram of a fish population prediction model provided by an embodiment of the present invention, wherein the trained fish population prediction model first predicts the number of plankton in each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the number of fish according to the predicted number of plankton, including: The plankton quantity is predicted according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted plankton quantity of each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; In this embodiment, the plankton quantity prediction is performed according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted number of plankton in each ocean layer, including: The real-time water environment data of each ocean layer is preprocessed through the input layer of the plankton prediction module to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; According to the real-time water environment data processed by each ocean layer, the final hidden state of the LSTM unit is calculated and output through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit in the first hidden layer of the plankton prediction module; Exemplarily, the calculation formulas for the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit are: The calculation formula of the input gate is: ;in, is the value of the input gate; is the sigmoid activation function; , and are the first weight matrix, the second weight matrix and the bias term of the input gate; The calculation formula of the forget gate is: ;in, is the value of the forget gate; is the sigmoid activation function; , and are the first weight matrix, the second weight matrix and the bias term of the forget gate; The calculation formula of the output gate is: ;in, is the value of the output gate; is the sigmoid activation function; , and are the first weight matrix, the second weight matrix and the bias term of the output gate; The calculation formula for candidate cell state calculation is: ;in, is the value calculated for the candidate cell state; tanh is the hyperbolic tangent activation function; , and are the first weight matrix, the second weight matrix and the bias term for calculating the candidate cell state; The calculation formula for cell state update is: ;in, The value updated for the cell state; is the cell state at the previous moment; Represents element-wise multiplication; is the value of the forget gate; is the value of the input gate; The values ​​calculated for the candidate cell states; The hidden state output is calculated as: ;in, The value output for the hidden state; is the value of the output gate; tanh is the hyperbolic tangent activation function; The value updated for the cell state; Represents element-wise multiplication; The hidden state of the LSTM unit is set as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit of the first hidden layer in the plankton prediction module; Exemplarily, the calculation formulas for the update gate, reset gate, candidate hidden state, and hidden state update of the GRU unit are: The calculation formula of the update gate is: ;in, represents the value of the update gate at time step t; is the sigmoid activation function; is the preset weight matrix of the update gate; Indicates that the hidden state of the previous moment And the input data of the GRU unit To splice; is the preset bias term of the update gate; The calculation formula for the reset gate is: ;in, is the value of the reset gate at time step t; is the sigmoid activation function; is the preset weight matrix of the reset gate; Indicates that the hidden state of the previous moment And the input data of the GRU unit To splice; is the preset bias term for the reset gate; The candidate hidden state is calculated as: ;in, is the candidate hidden state at time step t; tanh is the hyperbolic tangent activation function; is the value of the reset gate at time step t; and The preset weight matrix and bias term for the candidate hidden state; is element-wise multiplication; is the hidden state at the previous moment; Is the input data of the GRU unit; The calculation formula for hidden state update is: ;in, Update the final hidden state of the output for the hidden state; represents the value of the update gate at time step t; is the hidden state at the previous moment; is the candidate hidden state at time step t; is element-wise multiplication; According to the final hidden state of the GRU unit, the plankton quantity is predicted through the fully connected unit of the output layer in the plankton prediction module to obtain the predicted plankton quantity in each ocean layer.

[0027] According to the predicted number of plankton in each ocean layer and the real-time water environment data, the fish population is predicted through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

[0028] In this embodiment, the function of the adaptive weight layer is to assign weights to the predicted number of plankton in each ocean layer and the real-time water environment data; the second hidden layer is composed of a number of neurons, and its function is to predict the number of fish schools based on the weighted predicted number of plankton in each ocean layer and the real-time water environment data, so as to obtain the predicted number of fish schools in each ocean layer.

[0029] In this embodiment, the model training steps of the fish population prediction model are: Obtain historical water environment data on the actual number of hairtail fish stocks in each marine layer of the marine ranch; The plankton quantity is predicted according to the historical water environment data with actual quantity labels of fish shoals in each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module in the fish shoal quantity prediction model to be trained, so as to obtain the predicted quantity of plankton in each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; and the output layer includes a fully connected layer; According to the predicted number of plankton in each ocean layer and the historical water environment data, the fish population is predicted through the input layer, the adaptive weight layer, the second hidden layer and the output layer of the fish population prediction module in the fish population prediction model to be trained, so as to obtain the predicted number of fish populations in each ocean layer; According to the predicted number of fish in each ocean layer and the actual number of fish, the loss value is calculated through the loss function, and the model parameters of the fish population prediction model to be trained are optimized according to the loss value until the loss value converges to obtain a trained fish population prediction model; wherein the loss function includes but is not limited to the mean square error loss function, the mean absolute error loss function and the binary cross entropy loss function; the optimization process includes but is not limited to the stochastic gradient descent optimization algorithm and the Adam optimizer.

[0030] Step S4: Calculate the comprehensive number of fish schools in each ocean layer based on the number of fish schools identified and the predicted number of fish schools in each ocean layer, and calculate the total number of fish schools in the marine ranch to be counted based on the comprehensive number of fish schools in each ocean layer.

[0031] In this embodiment, the calculation formula for the comprehensive number of fish schools in each ocean layer is: ; in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks in the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ; in, The total number of marine ranch fish stocks to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

[0032] Embodiment 2: Reference Figure 2 , is a system structure diagram of a fishery resource calculation system for a marine ranch provided by an embodiment of the present invention, the system at least includes a data acquisition module, a fish population identification module, a fish population prediction module and a fish population total population calculation module; The data acquisition module is used to divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; The fish school number recognition module is used to input the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish schools recognized in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The fish population prediction module is used to input the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, so as to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; The total fish number calculation module is used to calculate the comprehensive number of fish schools in each ocean layer according to the number of fish school identification and the predicted number of fish schools in each ocean layer, and calculate the total number of fish schools to be counted in the marine ranch according to the comprehensive number of fish schools in each ocean layer.

[0033] In this embodiment, the fish school image recognition model includes: a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer of the local feature extraction module to obtain a local feature map of each ocean layer; The global feature extraction module extracts global features of the image from the local feature maps of each ocean layer through the multi-head autonomous intention layer and the feedforward neural network layer of the Transformer encoder in the global feature extraction module to obtain the global feature map of each ocean layer; The feature fusion module fuses the local feature map and the global feature map of each ocean layer respectively through the feature fusion module to obtain a fused feature map of each ocean layer; The YOLOv5 recognition module performs fish identification and quantity statistics on the fusion feature graph of each ocean layer through the backbone network layer, the neck network layer and the detection head layer of the YOLOv5 recognition module to obtain the number of fish schools identified in each ocean layer.

[0034] In this embodiment, the fish population prediction model includes: a plankton prediction module and a fish population prediction module; The plankton prediction module is used to predict the number of plankton according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module, so as to obtain the predicted number of plankton in each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; The fish population prediction module is used to predict the number of fish populations in each ocean layer based on the predicted number of plankton in each ocean layer and real-time water environment data, through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

[0035] In this embodiment, the plankton prediction module includes an input layer, a first hidden layer and an output layer; the first hidden layer includes an LSTM unit and a GRU unit; The input layer of the plankton prediction module is used to perform data preprocessing on the real-time water environment data of each ocean layer to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; The LSTM unit of the first hidden layer in the plankton prediction module is used to calculate and output the final hidden state of the LSTM unit through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit of the first hidden layer in the plankton prediction module according to the real-time water environment data processed by each ocean layer; The GRU unit of the first hidden layer in the plankton prediction module is used to set the hidden state of the LSTM unit as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit in the first hidden layer of the plankton prediction module; The output layer in the plankton prediction module is used to predict the number of plankton according to the final hidden state of the GRU unit through the fully connected units of the output layer in the plankton prediction module to obtain the predicted number of plankton in each ocean layer.

[0036] In this embodiment, the calculation formula for the comprehensive number of fish schools in each ocean layer is: ; in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks for the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ; in, The total number of marine ranch fish stocks to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

[0037] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for calculating fishery resources of marine ranches, characterized in that: include: Divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; Inputting the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish school identification in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; Inputting the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; Based on the number of fish schools identified and predicted in each ocean layer, the comprehensive number of fish schools in each ocean layer is calculated, and based on the comprehensive number of fish schools in each ocean layer, the total number of fish schools to be counted in the marine ranch is calculated.

2. The method for calculating fishery resources of marine ranches according to claim 1, characterized in that: The trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer to obtain the number of fish school identifications in each ocean layer, including: The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer to obtain the local feature map of each ocean layer; The global feature maps of each ocean layer are extracted by using the multi-head autonomous intention layer and feedforward neural network layer of the Transformer encoder in the global feature extraction module. The local feature map and the global feature map of each ocean layer are fused respectively through the feature fusion module to obtain the fused feature map of each ocean layer; The backbone network layer, neck network layer and detection head layer of the YOLOv5 recognition module are used to identify and count the fused feature maps of each ocean layer to obtain the number of fish schools identified in each ocean layer.

3. The method for calculating fishery resources of marine ranches according to claim 2, characterized in that: The trained fish population prediction model first predicts the number of plankton in each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the number of fish according to the predicted number of plankton, including: The plankton quantity is predicted according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted plankton quantity of each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; According to the predicted number of plankton in each ocean layer and the real-time water environment data, the fish population is predicted through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

4. The method for calculating fishery resources of marine ranches according to claim 3, characterized in that: The plankton quantity prediction is performed according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module to obtain the predicted number of plankton in each ocean layer, including: The real-time water environment data of each ocean layer is preprocessed through the input layer of the plankton prediction module to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; According to the real-time water environment data processed by each ocean layer, the final hidden state of the LSTM unit is calculated and output through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit in the first hidden layer of the plankton prediction module; The hidden state of the LSTM unit is set as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit of the first hidden layer in the plankton prediction module; According to the final hidden state of the GRU unit, the plankton quantity is predicted through the fully connected unit of the output layer in the plankton prediction module to obtain the predicted plankton quantity in each ocean layer.

5. The method for calculating fishery resources of marine ranches according to claim 4, characterized in that: The calculation formula for the comprehensive number of fish schools in each ocean layer is: ； in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks for the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ； in, The total number of marine ranch fish stocks to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

6. A fishery resource calculation system for marine ranches, characterized in that: include: Data acquisition module, fish population identification module, fish population prediction module and fish population total population calculation module; The data acquisition module is used to divide the marine ranch to be counted into several marine layers of different depth ranges, and collect real-time underwater sonar images and real-time water environment data of each marine layer; wherein the real-time water environment data includes light intensity, water temperature, water salt concentration and water nutrient concentration; The fish school number recognition module is used to input the real-time underwater sonar images of each ocean layer into the trained fish school image recognition model, so that the trained fish school image recognition model performs fish identification statistics on each ocean layer according to the real-time underwater sonar images of each ocean layer, and obtains the number of fish schools recognized in each ocean layer; wherein the fish school image recognition model includes a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The fish population prediction module is used to input the real-time water environment data of each ocean layer into the trained fish population prediction model, so that the trained fish population prediction model first predicts the plankton population of each ocean layer according to the real-time water environment data of each ocean layer, and then predicts the fish population according to the predicted plankton population, so as to obtain the predicted fish population of each ocean layer; wherein the fish population prediction model includes a plankton prediction module and a fish population prediction module; The total fish number calculation module is used to calculate the comprehensive number of fish schools in each ocean layer according to the number of fish school identification and the predicted number of fish schools in each ocean layer, and calculate the total number of fish schools to be counted in the marine ranch according to the comprehensive number of fish schools in each ocean layer.

7. The fishery resource calculation system of the marine ranch according to claim 6 is characterized in that: The fish school image recognition model includes: a local feature extraction module, a global feature extraction module, a feature fusion module and a YOLOv5 recognition module; The local feature extraction module extracts local features of the real-time underwater sonar images of each ocean layer through the convolution layer, pooling layer and activation function layer of the local feature extraction module to obtain a local feature map of each ocean layer; The global feature extraction module extracts global features of the image from the local feature maps of each ocean layer through the multi-head autonomous intention layer and the feedforward neural network layer of the Transformer encoder in the global feature extraction module to obtain the global feature map of each ocean layer; The feature fusion module fuses the local feature map and the global feature map of each ocean layer respectively through the feature fusion module to obtain a fused feature map of each ocean layer; The YOLOv5 recognition module performs fish identification and quantity statistics on the fusion feature graph of each ocean layer through the backbone network layer, the neck network layer and the detection head layer of the YOLOv5 recognition module to obtain the number of fish schools identified in each ocean layer.

8. The fishery resource calculation system of the marine ranch according to claim 7 is characterized in that: The fish population prediction model comprises: a plankton prediction module and a fish population prediction module; The plankton prediction module is used to predict the number of plankton according to the real-time water environment data of each ocean layer through the input layer, the first hidden layer and the output layer of the plankton prediction module, so as to obtain the predicted number of plankton in each ocean layer; wherein the first half of the first hidden layer is composed of a plurality of LSTM units, and the second half of the first hidden layer is composed of a plurality of GRU units; the output layer includes a fully connected layer; The fish population prediction module is used to predict the number of fish populations in each ocean layer based on the predicted number of plankton in each ocean layer and real-time water environment data, through the input layer, adaptive weight layer, second hidden layer and output layer of the fish population prediction module to obtain the predicted number of fish populations in each ocean layer.

9. The fishery resource calculation system of the marine ranch according to claim 8, characterized in that: The plankton prediction module includes an input layer, a first hidden layer and an output layer; the first hidden layer includes an LSTM unit and a GRU unit; The input layer of the plankton prediction module is used to perform data preprocessing on the real-time water environment data of each ocean layer to obtain the processed real-time water environment data of each ocean layer; wherein the data preprocessing includes data cleaning and normalization; The LSTM unit of the first hidden layer in the plankton prediction module is used to calculate and output the final hidden state of the LSTM unit through the input gate, forget gate, output gate, candidate cell state calculation, cell state update and hidden state output of the LSTM unit of the first hidden layer in the plankton prediction module according to the real-time water environment data processed by each ocean layer; The GRU unit of the first hidden layer in the plankton prediction module is used to set the hidden state of the LSTM unit as the initial hidden state of the GRU unit, and according to the real-time water environment data processed by each ocean layer and the initial hidden state of the GRU unit, the final hidden state of the GRU unit is calculated and output through the update gate, reset gate, candidate hidden state and hidden state update of the GRU unit in the first hidden layer of the plankton prediction module; The output layer in the plankton prediction module is used to predict the number of plankton according to the final hidden state of the GRU unit through the fully connected units of the output layer in the plankton prediction module to obtain the predicted number of plankton in each ocean layer.

10. The fishery resource calculation system of the marine ranch according to claim 9, characterized in that: The calculation formula for the comprehensive number of fish schools in each ocean layer is: ； in, is the total number of fish stocks in the ith ocean layer; is the number of fish identifications for the ith ocean layer; Predict the number of fish stocks in the ith ocean layer; and Preset weighting factors for the number of fish schools identified and the number of fish schools predicted, respectively; The calculation formula for the total number of fish stocks in the marine ranch to be counted is: ； in, The total number of marine ranch fish stocks to be counted; is the total number of fish schools in the i-th ocean layer; i=1, 2, 3, …, m, where m is a positive integer, i is the number of the ocean layer, and m is the total number of layers in which the ocean layer is divided.

Images