Marine water temperature-based phytoplankton density identification method and system

Abstract

The invention provides a phytoplankton density identification method and system based on ocean water temperature. The method comprises the following steps: acquiring a water temperature data set of a preset three-dimensional space of each grid region in a target ocean region within a preset time range; performing feature extraction on each water temperature data set to obtain a time sequence dynamic feature, a vertical layering feature and a space correlation feature of the ocean water temperature of each grid region; inputting the time sequence dynamic features, the vertical layering features and the spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on the time sequence dynamic features, the vertical layering features and the spatial correlation features based on an attention mechanism and a full-connection network to obtain fusion feature vectors, and generating the phytoplankton density of each grid region according to each fusion feature vector, thereby improving the accuracy and efficiency of phytoplankton density identification.

Description

Technical Field

[0001] The present application relates to the fields of phytoplankton density identification and machine learning technology, and in particular to a phytoplankton density identification method and system based on ocean water temperature. Background Art

[0002] As primary producers in marine ecosystems, phytoplankton density dynamics directly influence carbon cycling, red tide warnings, and biogeochemical cycles, and are key indicators in global climate change research. Traditional phytoplankton density monitoring relies primarily on manual microscopic counting or flow cytometry, which suffers from low efficiency, limited temporal and spatial coverage, and is difficult to meet the needs of real-time monitoring in large sea areas. Although rapid detection technologies such as spectral analysis and fluorescent labeling have developed in recent years, they are expensive and lack adaptability to complex environments. In particular, in multi-coupling scenarios such as dynamic water temperature and salinity stratification and nutrient gradients, it is difficult to accurately capture the nonlinear response characteristics of phytoplankton communities.

[0003] Although existing water temperature data inversion methods based on remote sensing or in-situ sensors can partially correlate with phytoplankton density, they face two major technical bottlenecks: First, traditional statistical models (such as Gaussian regression and logistic regression) have limited modeling capabilities for the dynamic characteristics of water temperature, and it is difficult to capture the complex relationship between vertical stratification parameters such as thermocline intensity and mixed layer depth (MLD) and density distribution; second, the cost of fine-grained annotation of phytoplankton density in massive monitoring data is extremely high, especially for water temperature series with high temporal and spatial resolution. Sparse annotation leads to a significant decrease in the generalization performance of supervised learning models, which cannot meet the needs of cross-regional migration and long-term ecological assessment.

[0004] In recent years, deep learning techniques have demonstrated potential in underwater image classification and ecological modeling. For example, the application of convolutional neural networks (CNNs) to phytoplankton morphology recognition has achieved high accuracy. In this context, the learning from label proportions (LLP) algorithm, using a weakly supervised paradigm, requires only regional density level labels, rather than individual sample labels, offering a new approach to reducing data labeling costs. Label proportion learning is a weakly supervised learning paradigm whose core principle is to train models based on the class ratios of a population of samples, rather than individual labels, significantly reducing data labeling costs. In LLP, data is divided into multiple "packages," each of which provides only a global distribution of class labels, without requiring specific labels for each sample. This learning approach is particularly well-suited for applications in areas such as ecological monitoring and remote sensing image analysis, where labeling is challenging and data volumes are large. Therefore, the label proportion learning algorithm could be considered for phytoplankton density recognition. Summary of the Invention

[0005] In response to the above technical problems, the present application provides a phytoplankton density identification method and system based on ocean water temperature to improve the accuracy and efficiency of phytoplankton density identification.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying phytoplankton density based on ocean water temperature, comprising: Obtaining a water temperature dataset of a preset three-dimensional space within a preset time range for each grid area in the target ocean area; Performing feature extraction on each of the water temperature data sets to obtain temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; Inputting each of the temporal dynamic features, vertical stratification features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical stratification features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generating the phytoplankton density of each of the grid areas based on each of the fused feature vectors; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

[0007] The embodiment of the present application provides a phytoplankton density identification method based on ocean water temperature. By collecting water temperature data at various locations and time points in the target ocean area, a water temperature data set is constructed, and then temporal dynamic features, vertical stratification features, and spatial correlation features are extracted from the water temperature data set and each feature is identified through a recognition model to generate phytoplankton density in different grid areas in the target ocean area. The embodiment of the present application comprehensively captures the complex relationship between ocean water temperature and phytoplankton density by integrating temporal dynamics, vertical stratification, and spatial correlation multidimensional features, significantly improving recognition accuracy. The embodiment of the present application also divides the target ocean area into multiple grid areas in advance, and then performs phytoplankton density identification on different grid areas respectively. This gridded data processing enables the embodiment of the present application to support high-resolution refined identification, meet the real-time monitoring needs of large-scale sea areas, and also better adapt to the label ratio learning algorithm. In addition, the present application introduces a label ratio learning algorithm based on the density characteristics of the recognition task, and combines it with a deep learning model to construct a recognition model, which greatly reduces the dependence on refined labeled data, reduces data labeling costs, and enhances the generalization ability of the model in sparsely labeled scenarios, thereby improving the accuracy and efficiency of phytoplankton density identification.

[0008] Furthermore, the step of obtaining a water temperature dataset of a preset three-dimensional space within a preset time range for each grid area in the target ocean area includes: Extracting sea surface temperature datasets for each of the preset three-dimensional spaces within a preset time range from satellite remote sensing data; Acquiring a stratified water temperature data set in each of the preset three-dimensional spaces within a preset time range through a plurality of preset sensors; By combining the sea surface temperature datasets and the layered water temperature datasets, a water temperature dataset of the preset three-dimensional space of each grid area within a preset time range is constructed.

[0009] The embodiment of the present application provides a method for acquiring a water temperature dataset. By combining satellite remote sensing and multiple preset sensors for data collection, the method can fully collect water temperature data at different locations within each grid area, obtaining multi-dimensional, multi-scale water temperature information covering the ocean surface and vertical space. By complementary fusion of sea surface and layered data, the integrity and representativeness of the water temperature dataset are improved, laying a reliable foundation for subsequent feature extraction and improving the accuracy of phytoplankton density identification. In addition, the embodiment of the present application uses both sensor technology and remote sensing technology to collect water temperature data, realizing the simultaneous collection of multiple data sources, avoiding the limitations and data errors of a single data source, and laying a reliable foundation for subsequent feature extraction.

[0010] In one possible implementation, extracting features from each of the water temperature datasets to obtain the temporal dynamic features of the ocean water temperature in each grid area includes: Performing data statistics on each of the water temperature data sets using a time sliding window of a preset size, and calculating the mean, standard deviation, and gradient change rate of the water temperature in each grid area within each time period; Based on the time series decomposition algorithm, each of the water temperature data sets is decomposed into time series to obtain the long-term trend characteristics, seasonal cycle characteristics and residual terms of the water temperature changes in each grid area; By combining the water temperature mean, standard deviation, gradient change rate, long-term trend characteristics, seasonal cycle characteristics and residual terms, the time series dynamic characteristics of the ocean water temperature in each grid area are constructed.

[0011] The present embodiment provides a method for extracting temporal dynamic features. This method effectively captures the short-term fluctuation characteristics of ocean water temperature by calculating the mean, standard deviation, and gradient change rate of water temperature using a data statistics method based on a time sliding window. The method also decomposes the water temperature dataset into temporal trend characteristics, seasonal cycle characteristics, and residual terms using a time series decomposition algorithm, enhancing the model's ability to analyze dynamic changes in water temperature and improving the temporal adaptability of density prediction. Finally, the short-term and long-term features are combined to construct temporal dynamic features of ocean water temperature in each grid area. This allows the model to accurately capture the temporal variation of water temperature, and then maps the temperature variation pattern to the phytoplankton density variation pattern, improving the accuracy of phytoplankton density identification.

[0012] In one possible implementation, extracting features from each of the water temperature datasets to obtain vertically layered features of the ocean water temperature in each grid area includes: Calculate and determine the mixed layer depth and thermocline depth of each grid area according to each water temperature data set; performing a gradient integral calculation on the water temperature data between the depth of the mixed layer and the depth of the thermocline according to each of the water temperature data sets, to obtain an integrated temperature gradient within a corresponding range for each grid area; The vertical stratification characteristics of the ocean water temperature in each grid area are constructed by combining the depth of the mixed layer, the depth of the thermocline and the integrated temperature gradient.

[0013] The embodiment of the present application provides a method for extracting vertical stratification characteristics. By calculating the mixing layer depth, thermocline depth and integrated temperature gradient, the physical characteristics of vertical water temperature stratification are quantified, so that the model can accurately capture the variation of water temperature with space, and then accurately characterize the regulatory effect of thermocline intensity on the vertical distribution of phytoplankton, solve the problem of insufficient modeling of stratification parameters in traditional models, and improve the accuracy of phytoplankton density identification.

[0014] In one possible implementation, extracting features from each of the water temperature datasets to obtain spatial correlation features of the ocean water temperature in each grid area includes: generating a four-dimensional continuous water temperature field including longitude, latitude, depth and time according to each of the water temperature data sets; Calculating the longitudinal water temperature gradient, latitudinal water temperature gradient, vertical water temperature gradient and temporal water temperature gradient of each sampling point of the four-dimensional continuous water temperature field according to each of the water temperature data sets; According to each of the longitudinal water temperature gradients, latitudinal water temperature gradients, vertical water temperature gradients and temporal water temperature gradients, a graph convolutional network model is performed on the target ocean area to obtain the spatial correlation characteristics of the ocean water temperature in each grid area.

[0015] An embodiment of the present application provides a method for extracting spatial correlation features. Taking into account the flow characteristics of the ocean, the water temperature change characteristics of the same grid area are not only affected by the time and space of the area, but also by the surrounding grid areas. Therefore, the embodiment of the present application constructs an overall four-dimensional continuous water temperature field and combines it with a graph convolutional network for modeling, effectively capturing the spatial correlation of longitude, latitude, depth and time dimensions, revealing the local and global patterns of water temperature changes, enhancing the model's adaptability to complex ocean spatial structures, and thereby improving the accuracy of phytoplankton density identification.

[0016] Furthermore, the recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

[0017] The embodiment of the present application provides a phytoplankton density identification method based on a recognition model. The input features are converted into corresponding feature vectors through feature encoding and multi-scale pooling. The feature vectors are then weighted and fused based on the attention mechanism. Finally, the phytoplankton density of each grid area is generated based on the fused feature vectors. The embodiment of the present application dynamically assigns weights to different features through the attention mechanism, focuses on key information, suppresses noise interference, and optimizes the feature fusion effect. The fully convolutional network is combined with the LeakyReLU activation function to improve the nonlinear expression ability of the model, enhance the modeling ability of the nonlinear response characteristics of phytoplankton density, and improve the accuracy and efficiency of phytoplankton density identification.

[0018] In one possible implementation, the initial recognition model is trained based on a plurality of historical phytoplankton density data and a plurality of historical water temperature data sets of the target ocean area at different periods to obtain the recognition model, including: The initial recognition model is obtained based on a deep learning model, wherein the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer, and an output layer; According to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets, dividing the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area; Performing feature extraction on each of the training water temperature data sets to obtain each corresponding training water temperature feature set; According to each of the training phytoplankton density data and each of the training water temperature feature sets, a plurality of label ratio training data packets are created for each grid area, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The initial recognition model is trained according to a preset hybrid loss function and each of the label ratio training data packets to determine the deep learning parameters and the label ratio learning parameters, and then the recognition model is updated, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the ratio label.

[0019] The embodiment of the present application provides a method for training a recognition model. According to the characteristics of the phytoplankton density recognition task, the water temperature of each location point cannot reflect the phytoplankton density of the entire grid area, and the phytoplankton density of the entire grid area is equivalent to the proportional label of the package in the label ratio learning. Therefore, the recognition task can be regarded as a label ratio learning task, and then the initial recognition model is trained based on the label ratio learning algorithm. After the label ratio learning algorithm is introduced, according to the characteristics of the weak supervision paradigm of the label ratio learning algorithm, only the regional density ratio label is required to train the model without labeling each data, which significantly reduces the labeling cost and solves the sparse labeling problem. In addition, the embodiment of the present application introduces the Wasserstein distance loss in the hybrid loss function. The Wasserstein distance measures the distribution difference through the optimal transmission theory. Even if the support set of the true label distribution and the predicted distribution does not overlap, it can still provide an effective gradient signal. In the LLP-based model process, when the category distribution of multiple label packages is quite different, the Wasserstein distance can avoid training stagnation caused by distribution separation and improve the robustness of the model.

[0020] In a second aspect, an embodiment of the present application provides a phytoplankton density identification system based on ocean water temperature, comprising an acquisition module, a feature extraction module, and an identification module; The acquisition module is used to acquire a water temperature dataset of a preset three-dimensional space within a preset time range in each grid area in the target ocean area; The feature extraction module is used to extract features from each of the water temperature data sets to obtain the temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; The recognition module is used to input each of the temporal dynamic features, vertical layered features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical layered features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generate the phytoplankton density of each grid area based on each fused feature vector; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

[0021] Furthermore, the recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

[0022] In one possible implementation, the phytoplankton density identification system further includes a model training module, which is used to train an initial identification model based on a number of historical phytoplankton density data and a number of historical water temperature data sets of a target ocean area at different periods to obtain the identification model, and includes a model construction unit, a training data set division unit, a feature extraction unit, a training data construction unit, and a training unit; The model construction unit is used to obtain the initial recognition model based on the deep learning model, and the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer and an output layer; The training data set division unit is used to divide the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area according to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets; The feature extraction unit is used to extract features from each of the training water temperature data sets to obtain each corresponding training water temperature feature set; The training data construction unit is used to create a plurality of label ratio training data packets for each grid area according to each of the training phytoplankton density data and each of the training water temperature feature sets, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The training unit is used to train the initial recognition model according to a preset hybrid loss function and each of the label ratio training data packets, determine the deep learning parameters and the label ratio learning parameters, and then update the recognition model, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the proportional label. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic flow chart of a method for identifying phytoplankton density based on ocean water temperature provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an identification model in a method for identifying phytoplankton density based on ocean water temperature provided in an embodiment of the present application; Figure 3 A schematic structural diagram of a phytoplankton density identification system based on ocean water temperature provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] It should be noted that the step numbers herein are for convenience of explanation of the specific embodiments and do not serve to define the order in which the steps are to be performed. In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature designated "first" or "second" may explicitly or implicitly include one or more of such features.

[0026] Example 1: like Figure 1 As shown, embodiment 1 provides a method for identifying phytoplankton density based on ocean water temperature, including steps S1 to S3: Step S1: obtaining a water temperature dataset of a preset three-dimensional space within a preset time range in each grid area in the target ocean area; Step S2: extracting features from each of the water temperature data sets to obtain temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; Step S3, inputting each of the temporal dynamic features, vertical layered features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical layered features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generating the phytoplankton density of each of the grid areas according to each of the fused feature vectors; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

[0027] The embodiment of the present application provides a phytoplankton density identification method based on ocean water temperature. By collecting water temperature data at various locations and time points in the target ocean area, a water temperature data set is constructed, and then temporal dynamic features, vertical stratification features, and spatial correlation features are extracted from the water temperature data set and each feature is identified through a recognition model to generate phytoplankton density in different grid areas in the target ocean area. The embodiment of the present application comprehensively captures the complex relationship between ocean water temperature and phytoplankton density by integrating temporal dynamics, vertical stratification, and spatial correlation multidimensional features, significantly improving recognition accuracy. The embodiment of the present application also divides the target ocean area into multiple grid areas in advance, and then performs phytoplankton density identification on different grid areas respectively. This gridded data processing enables the embodiment of the present application to support high-resolution refined identification, meet the real-time monitoring needs of large-scale sea areas, and also better adapt to the label ratio learning algorithm. In addition, the present application introduces a label ratio learning algorithm based on the density characteristics of the recognition task, and combines it with a deep learning model to construct a recognition model, which greatly reduces the dependence on refined labeled data, reduces data labeling costs, and enhances the generalization ability of the model in sparsely labeled scenarios, thereby improving the accuracy and efficiency of phytoplankton density identification.

[0028] Furthermore, in step S1, obtaining a water temperature dataset of a preset three-dimensional space within a preset time range for each grid area in the target ocean area includes: Extracting sea surface temperature datasets for each of the preset three-dimensional spaces within a preset time range from satellite remote sensing data; Acquiring a stratified water temperature data set in each of the preset three-dimensional spaces within a preset time range through a plurality of preset sensors; By combining the sea surface temperature datasets and the layered water temperature datasets, a water temperature dataset of the preset three-dimensional space of each grid area within a preset time range is constructed.

[0029] The embodiment of the present application provides a method for acquiring a water temperature dataset. By combining satellite remote sensing and multiple preset sensors for data collection, the method can fully collect water temperature data at different locations within each grid area, obtaining multi-dimensional, multi-scale water temperature information covering the ocean surface and vertical space. By complementary fusion of sea surface and layered data, the integrity and representativeness of the water temperature dataset are improved, laying a reliable foundation for subsequent feature extraction and improving the accuracy of phytoplankton density identification. In addition, the embodiment of the present application uses both sensor technology and remote sensing technology to collect water temperature data, realizing the simultaneous collection of multiple data sources, avoiding the limitations and data errors of a single data source, and laying a reliable foundation for subsequent feature extraction.

[0030] In a preferred embodiment, before density identification, the target ocean area is divided into several three-dimensional grids based on longitude, latitude, and depth. The size of each grid can be set according to the required monitoring resolution. After gridding, each grid is assigned a unique identifier, resulting in several grid regions. It is important to note that each grid contains the sea surface. During the water temperature data acquisition process, a sea surface temperature (SST) dataset for the target area is acquired from MODIS or VIIRS satellite sensors, covering the past year at a daily resolution. This yields a SST dataset for each grid. Argo float sensors deployed at different depths in the target ocean area collect vertically stratified water temperature data within the grid region, for example, from 0 to 200 meters, with one Argo float deployed every 10 meters. The temporal range is synchronized with the satellite data, resulting in a stratified water temperature dataset for each grid. The two-dimensional SST data from the satellite remote sensing is aligned with the three-dimensional stratified water temperature data from the float sensors. Missing values ​​are filled using an interpolation algorithm to generate a spatiotemporally continuous water temperature dataset for each grid, formatted as a four-dimensional tensor (longitude, latitude, depth, time).

[0031] In one possible implementation, in step S2, the feature extraction of each of the water temperature data sets to obtain the temporal dynamic characteristics of the ocean water temperature in each grid area includes: Performing data statistics on each of the water temperature data sets using a time sliding window of a preset size, and calculating the mean, standard deviation, and gradient change rate of the water temperature in each grid area within each time period; Based on the time series decomposition algorithm, each of the water temperature data sets is decomposed into time series to obtain the long-term trend characteristics, seasonal cycle characteristics and residual terms of the water temperature changes in each grid area; By combining the water temperature mean, standard deviation, gradient change rate, long-term trend characteristics, seasonal cycle characteristics and residual terms, the time series dynamic characteristics of the ocean water temperature in each grid area are constructed.

[0032] The present embodiment provides a method for extracting temporal dynamic features. This method effectively captures the short-term fluctuation characteristics of ocean water temperature by calculating the mean, standard deviation, and gradient change rate of water temperature using a data statistics method based on a time sliding window. The method also decomposes the water temperature dataset into temporal trend characteristics, seasonal cycle characteristics, and residual terms using a time series decomposition algorithm, enhancing the model's ability to analyze dynamic changes in water temperature and improving the temporal adaptability of density prediction. Finally, the short-term and long-term features are combined to construct temporal dynamic features of ocean water temperature in each grid area. This allows the model to accurately capture the temporal variation of water temperature, and then maps the temperature variation pattern to the phytoplankton density variation pattern, improving the accuracy of phytoplankton density identification.

[0033] In a preferred embodiment, when extracting the time series dynamic characteristics of ocean water temperature, the original water temperature data set is smoothed by a Savitzky-Golay filter to retain the long-term trend while suppressing short-term disturbances. Then a 30-day time window is set with a sliding step of 7 days. The mean, standard deviation and gradient change rate within the window are calculated for the water temperature data set of each grid. In the time series decomposition process, Fourier series constraints are introduced in the seasonal component to solve the defect of traditional STL being sensitive to non-integer periods. For example, for the composite pattern of an annual cycle (365 days) superimposed on a monthly cycle (30 days), frequency domain filtering is used to separate periodic terms of different scales. In addition, an autocorrelation test (ACF / PACF) is performed on the residual term after decomposition to identify mutation events not captured by the model and use them as abnormal event feature vectors.

[0034] In one possible implementation, in step S2, the feature extraction of each of the water temperature data sets to obtain vertical layered features of the ocean water temperature in each grid area includes: Calculate and determine the mixed layer depth and thermocline depth of each grid area according to each water temperature data set; performing a gradient integral calculation on the water temperature data between the depth of the mixed layer and the depth of the thermocline according to each of the water temperature data sets, to obtain an integrated temperature gradient within a corresponding range for each grid area; The vertical stratification characteristics of the ocean water temperature in each grid area are constructed by combining the depth of the mixed layer, the depth of the thermocline and the integrated temperature gradient.

[0035] The embodiment of the present application provides a method for extracting vertical stratification characteristics. By calculating the mixing layer depth, thermocline depth and integrated temperature gradient, the physical characteristics of vertical water temperature stratification are quantified, so that the model can accurately capture the variation of water temperature with space, and then accurately characterize the regulatory effect of thermocline intensity on the vertical distribution of phytoplankton, solve the problem of insufficient modeling of stratification parameters in traditional models, and improve the accuracy of phytoplankton density identification.

[0036] In a preferred embodiment, when extracting the vertical stratification characteristics of ocean water temperature, the mixed layer depth of each grid area is determined based on the water temperature dataset using a density change threshold method. Simultaneously, the water temperature gradient distribution of each grid area is calculated, and curvature analysis is used to locate the depth layer with the maximum water temperature gradient in each grid area, thereby determining the thermocline depth of each grid area. Finally, the vertical water temperature gradient between the mixed layer depth and the thermocline depth in each grid area is normalized and integrated. This indicator can be used to represent the degree of resistance of the thermocline to the vertical migration of phytoplankton.

[0037] In one possible implementation, in step S2, extracting features from each of the water temperature datasets to obtain spatial correlation features of the ocean water temperature in each grid area includes: generating a four-dimensional continuous water temperature field including longitude, latitude, depth and time according to each of the water temperature data sets; Calculating the longitudinal water temperature gradient, latitudinal water temperature gradient, vertical water temperature gradient and temporal water temperature gradient of each sampling point of the four-dimensional continuous water temperature field according to each of the water temperature data sets; According to each of the longitudinal water temperature gradients, latitudinal water temperature gradients, vertical water temperature gradients and temporal water temperature gradients, a graph convolutional network model is performed on the target ocean area to obtain the spatial correlation characteristics of the ocean water temperature in each grid area.

[0038] An embodiment of the present application provides a method for extracting spatial correlation features. Taking into account the flow characteristics of the ocean, the water temperature change characteristics of the same grid area are not only affected by the time and space of the area, but also by the surrounding grid areas. Therefore, the embodiment of the present application constructs an overall four-dimensional continuous water temperature field and combines it with a graph convolutional network for modeling, effectively capturing the spatial correlation of longitude, latitude, depth and time dimensions, revealing the local and global patterns of water temperature changes, enhancing the model's adaptability to complex ocean spatial structures, and thereby improving the accuracy of phytoplankton density identification.

[0039] In a preferred embodiment, when extracting the spatial correlation characteristics of ocean water temperature, the Sobel operator is used to calculate the longitudinal, latitudinal, vertical, and temporal gradients of the four-dimensional continuous water temperature field based on each water temperature dataset, forming a 32-dimensional gradient tensor. Covariance matrix analysis is then performed to quantify the principal component analysis (PCA) of the gradient direction, extracting the dominant transmission direction characteristics and obtaining the longitudinal, latitudinal, vertical, and temporal water temperature gradients at each sampling point. Finally, a graph structure is constructed using grids as nodes and gradients as edge weights. Neighborhood information is aggregated using a graph convolutional network (GCN), outputting a spatial correlation feature vector for each grid.

[0040] Furthermore, in step S3, the recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

[0041] The embodiment of the present application provides a phytoplankton density identification method based on a recognition model. The input features are converted into corresponding feature vectors through feature encoding and multi-scale pooling. The feature vectors are then weighted and fused based on the attention mechanism. Finally, the phytoplankton density of each grid area is generated based on the fused feature vectors. The embodiment of the present application dynamically assigns weights to different features through the attention mechanism, focuses on key information, suppresses noise interference, and optimizes the feature fusion effect. The fully convolutional network is combined with the LeakyReLU activation function to improve the nonlinear expression ability of the model, enhance the modeling ability of the nonlinear response characteristics of phytoplankton density, and improve the accuracy and efficiency of phytoplankton density identification.

[0042] In one possible implementation, the initial recognition model is trained based on a plurality of historical phytoplankton density data and a plurality of historical water temperature data sets of the target ocean area at different periods to obtain the recognition model, including: The initial recognition model is obtained based on a deep learning model, wherein the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer, and an output layer; According to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets, dividing the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area; Performing feature extraction on each of the training water temperature data sets to obtain each corresponding training water temperature feature set; According to each of the training phytoplankton density data and each of the training water temperature feature sets, a plurality of label ratio training data packets are created for each grid area, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The initial recognition model is trained according to a preset hybrid loss function and each of the label ratio training data packets to determine the deep learning parameters and the label ratio learning parameters, and then the recognition model is updated, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the ratio label.

[0043] The embodiment of the present application provides a method for training a recognition model. According to the characteristics of the phytoplankton density recognition task, the water temperature of each location point cannot reflect the phytoplankton density of the entire grid area, and the phytoplankton density of the entire grid area is equivalent to the proportional label of the package in the label ratio learning. Therefore, the recognition task can be regarded as a label ratio learning task, and then the initial recognition model is trained based on the label ratio learning algorithm. After the label ratio learning algorithm is introduced, according to the characteristics of the weak supervision paradigm of the label ratio learning algorithm, only the regional density ratio label is required to train the model without labeling each data, which significantly reduces the labeling cost and solves the sparse labeling problem. In addition, the embodiment of the present application introduces the Wasserstein distance loss in the hybrid loss function. The Wasserstein distance measures the distribution difference through the optimal transmission theory. Even if the support set of the true label distribution and the predicted distribution does not overlap, it can still provide an effective gradient signal. In the LLP-based model process, when the category distribution of multiple label packages is quite different, the Wasserstein distance can avoid training stagnation caused by distribution separation and improve the robustness of the model.

[0044] In a preferred embodiment, Figure 2 As shown, the initial recognition model comprises an input layer, an encoding layer, a pooling layer, a fully connected layer, a convolutional layer and an output layer, wherein the input layer is used to receive the temporal dynamic features, vertical hierarchical features and spatial correlation features of the external input; the encoding layer is used to process the temporal dynamics and vertical hierarchical features through a bidirectional gated recurrent unit (BiGRU), and output a 128-dimensional temporal feature vector and a vertical feature vector; the pooling layer is used to extract the multi-resolution information of the spatial correlation features through multi-scale pooling operations (maximum pooling, average pooling), and output a 128-dimensional spatial feature vector; the fully connected layer is used to calculate the attention weight of each feature vector, and input the attention weight and each feature vector into the convolutional layer; the convolutional layer is used to perform weighted fusion of each feature vector according to each attention weight through a fully convolutional network FCN and a LeakyReLU activation function; the output layer is used to output the proportion of phytoplankton in each grid area through a Softmax function combined with a label ratio parameter, and convert it into the corresponding phytoplankton density. When training the initial recognition model, the temporal, vertical, and spatial water temperature characteristics of each grid cell and the corresponding phytoplankton fraction are combined into a label-proportional training data package. The phytoplankton fraction is calculated by normalizing the actual phytoplankton density in the area. This label-proportional training data package is then fed into the initial recognition model, causing it to output the corresponding prediction results. The difference between the prediction results and the package labels is calculated using cross-entropy loss and Wasserstein distance, and the model parameters are then updated. Based on this training process, the Adam optimizer is used with a learning rate of 0.001 and a batch size of 32. Training is iterated until the loss converges to obtain the final recognition model.

[0045] Furthermore, the embodiment of the present application can also map the phytoplankton density to the corresponding phytoplankton density level according to a preset mapping relationship table, input the phytoplankton density level of each grid area into the geographic information system (GIS), generate a phytoplankton distribution heat map of the target ocean area, and realize the visualization of the density identification results. Since the embodiment of the present application can identify the phytoplankton density in real time, it can continuously generate phytoplankton distribution heat maps at different time points over time, revealing the changing patterns of marine phytoplankton and providing data support for subsequent ocean research work.

[0046] Example 2: like Figure 3 As shown, embodiment 2 provides a phytoplankton density identification system based on ocean water temperature, comprising an acquisition module 10, a feature extraction module 20, and an identification module 30; The acquisition module 10 is used to acquire a water temperature dataset of a preset three-dimensional space within a preset time range in each grid area in the target ocean area; The feature extraction module 20 is used to extract features from each of the water temperature data sets to obtain the temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; The recognition module 30 is used to input each of the temporal dynamic features, vertical layered features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical layered features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area based on each fused feature vector; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

[0047] Furthermore, the acquisition module 10 acquires a water temperature dataset of a preset three-dimensional space within a preset time range for each grid area in the target ocean area, including: Extracting sea surface temperature datasets for each of the preset three-dimensional spaces within a preset time range from satellite remote sensing data; Acquiring a stratified water temperature data set in each of the preset three-dimensional spaces within a preset time range through a plurality of preset sensors; By combining the sea surface temperature datasets and the layered water temperature datasets, a water temperature dataset of the preset three-dimensional space of each grid area within a preset time range is constructed.

[0048] In one possible implementation, the feature extraction module 20 performs feature extraction on each of the water temperature data sets to obtain the temporal dynamic features of the ocean water temperature in each grid area, including: Performing data statistics on each of the water temperature data sets using a time sliding window of a preset size, and calculating the mean, standard deviation, and gradient change rate of the water temperature in each grid area within each time period; Based on the time series decomposition algorithm, each of the water temperature data sets is decomposed into time series to obtain the long-term trend characteristics, seasonal cycle characteristics and residual terms of the water temperature changes in each grid area; By combining the water temperature mean, standard deviation, gradient change rate, long-term trend characteristics, seasonal cycle characteristics and residual terms, the time series dynamic characteristics of the ocean water temperature in each grid area are constructed.

[0049] In one possible implementation, the feature extraction module 20 performs feature extraction on each of the water temperature datasets to obtain vertical layered features of the ocean water temperature in each grid area, including: Calculate and determine the mixed layer depth and thermocline depth of each grid area according to each water temperature data set; performing a gradient integral calculation on the water temperature data between the depth of the mixed layer and the depth of the thermocline according to each of the water temperature data sets, to obtain an integrated temperature gradient within a corresponding range for each grid area; The vertical stratification characteristics of the ocean water temperature in each grid area are constructed by combining the depth of the mixed layer, the depth of the thermocline and the integrated temperature gradient.

[0050] In one possible implementation, the feature extraction module 20 performs feature extraction on each of the water temperature datasets to obtain spatial correlation features of the ocean water temperature in each grid area, including: generating a four-dimensional continuous water temperature field including longitude, latitude, depth and time according to each of the water temperature data sets; Calculating the longitudinal water temperature gradient, latitudinal water temperature gradient, vertical water temperature gradient and temporal water temperature gradient of each sampling point of the four-dimensional continuous water temperature field according to each of the water temperature data sets; According to each of the longitudinal water temperature gradients, latitudinal water temperature gradients, vertical water temperature gradients and temporal water temperature gradients, a graph convolutional network model is performed on the target ocean area to obtain the spatial correlation characteristics of the ocean water temperature in each grid area.

[0051] Furthermore, the recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

[0052] In one possible implementation, the phytoplankton density identification system further includes a model training module, which is used to train an initial identification model based on a number of historical phytoplankton density data and a number of historical water temperature data sets of a target ocean area at different periods to obtain the identification model, and includes a model construction unit, a training data set division unit, a feature extraction unit, a training data construction unit, and a training unit; The model construction unit is used to obtain the initial recognition model based on the deep learning model, and the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer and an output layer; The training data set division unit is used to divide the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area according to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets; The feature extraction unit is used to extract features from each of the training water temperature data sets to obtain each corresponding training water temperature feature set; The training data construction unit is used to create a plurality of label ratio training data packets for each grid area according to each of the training phytoplankton density data and each of the training water temperature feature sets, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The training unit is used to train the initial recognition model according to a preset hybrid loss function and each of the label ratio training data packets, determine the deep learning parameters and the label ratio learning parameters, and then update the recognition model, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the proportional label.

[0053] The embodiment of the present application provides a phytoplankton density identification system based on ocean water temperature. By collecting water temperature data at various locations and time points in the target ocean area, a water temperature data set is constructed. Then, temporal dynamic features, vertical stratification features, and spatial correlation features are extracted from the water temperature data set and each feature is identified through a recognition model to generate phytoplankton density in different grid areas in the target ocean area. The embodiment of the present application comprehensively captures the complex relationship between ocean water temperature and phytoplankton density by integrating temporal dynamics, vertical stratification, and spatial correlation multidimensional features, significantly improving recognition accuracy. The embodiment of the present application also divides the target ocean area into multiple grid areas in advance, and then performs phytoplankton density identification on different grid areas respectively. This gridded data processing enables the embodiment of the present application to support high-resolution refined identification, meet the real-time monitoring needs of large-scale sea areas, and also better adapt to the label ratio learning algorithm. In addition, the present application introduces a label ratio learning algorithm based on the density characteristics of the recognition task, and combines it with a deep learning model to build a recognition model, greatly reducing the dependence on refined labeled data, reducing data labeling costs, while enhancing the generalization ability of the model in sparsely labeled scenarios, and improving the accuracy and efficiency of phytoplankton density identification.

[0054] The more detailed working principle and process flow of this embodiment can be referred to, but not limited to, the relevant records of the first embodiment.

[0055] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application by those skilled in the art should be included within the scope of protection of this application.

Claims

1. A method for identifying phytoplankton density based on ocean water temperature, characterized in that: include: Obtaining a water temperature dataset of a preset three-dimensional space within a preset time range for each grid area in the target ocean area; Performing feature extraction on each of the water temperature data sets to obtain temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; Inputting each of the temporal dynamic features, vertical stratification features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical stratification features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generating the phytoplankton density of each of the grid areas based on each of the fused feature vectors; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

2. The method for identifying phytoplankton density based on ocean water temperature according to claim 1, wherein: The step of obtaining a water temperature dataset of a preset three-dimensional space within a preset time range in each grid area in the target ocean area includes: Extracting sea surface temperature datasets for each of the preset three-dimensional spaces within a preset time range from satellite remote sensing data; Acquiring a stratified water temperature data set in each of the preset three-dimensional spaces within a preset time range through a plurality of preset sensors; By combining the sea surface temperature datasets and the layered water temperature datasets, a water temperature dataset of the preset three-dimensional space of each grid area within a preset time range is constructed.

3. The method for identifying phytoplankton density based on ocean water temperature according to claim 1, wherein: The feature extraction of each of the water temperature data sets to obtain the temporal dynamic characteristics of the ocean water temperature in each grid area includes: Performing data statistics on each of the water temperature data sets using a time sliding window of a preset size, and calculating the mean, standard deviation, and gradient change rate of the water temperature in each grid area within each time period; Based on the time series decomposition algorithm, each of the water temperature data sets is decomposed into time series to obtain the long-term trend characteristics, seasonal cycle characteristics and residual terms of the water temperature changes in each grid area; By combining the water temperature mean, standard deviation, gradient change rate, long-term trend characteristics, seasonal cycle characteristics and residual terms, the time series dynamic characteristics of the ocean water temperature in each grid area are constructed.

4. The method for identifying phytoplankton density based on ocean water temperature according to claim 1, wherein: The feature extraction of each of the water temperature data sets to obtain vertical stratified features of the ocean water temperature in each grid area includes: Calculate and determine the mixed layer depth and thermocline depth of each grid area according to each water temperature data set; performing a gradient integral calculation on the water temperature data between the depth of the mixed layer and the depth of the thermocline according to each of the water temperature data sets, to obtain an integrated temperature gradient within a corresponding range for each grid area; The vertical stratification characteristics of the ocean water temperature in each grid area are constructed by combining the depth of the mixed layer, the depth of the thermocline and the integrated temperature gradient.

5. The method for identifying phytoplankton density based on ocean water temperature according to claim 1, wherein: The feature extraction of each of the water temperature data sets to obtain the spatial correlation features of the ocean water temperature in each grid area includes: generating a four-dimensional continuous water temperature field including longitude, latitude, depth and time according to each of the water temperature data sets; Calculating the longitudinal water temperature gradient, latitudinal water temperature gradient, vertical water temperature gradient and temporal water temperature gradient of each sampling point of the four-dimensional continuous water temperature field according to each of the water temperature data sets; According to each of the longitudinal water temperature gradients, latitudinal water temperature gradients, vertical water temperature gradients and temporal water temperature gradients, a graph convolutional network model is performed on the target ocean area to obtain the spatial correlation characteristics of the ocean water temperature in each grid area.

6. The method for identifying phytoplankton density based on ocean water temperature according to claim 1, wherein: The recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

7. A method for identifying phytoplankton density based on ocean water temperature according to any one of claims 1 to 6, characterized in that: The initial recognition model is trained based on a plurality of historical phytoplankton density data and a plurality of historical water temperature data sets of the target ocean area at different periods to obtain the recognition model, including: The initial recognition model is obtained based on a deep learning model, wherein the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer, and an output layer; According to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets, dividing the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area; Performing feature extraction on each of the training water temperature data sets to obtain each corresponding training water temperature feature set; According to each of the training phytoplankton density data and each of the training water temperature feature sets, a plurality of label ratio training data packets are created for each grid area, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The initial recognition model is trained according to a preset hybrid loss function and each of the label ratio training data packets to determine the deep learning parameters and the label ratio learning parameters, and then the recognition model is updated, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the ratio label.

8. A phytoplankton density identification system based on ocean water temperature, characterized in that: It includes an acquisition module, a feature extraction module and a recognition module; The acquisition module is used to acquire a water temperature dataset of a preset three-dimensional space within a preset time range in each grid area in the target ocean area; The feature extraction module is used to extract features from each of the water temperature data sets to obtain the temporal dynamic features, vertical stratification features, and spatial correlation features of the ocean water temperature in each grid area; The recognition module is used to input each of the temporal dynamic features, vertical layered features, and spatial correlation features into a preset recognition model, so that the recognition model performs feature fusion on each of the temporal dynamic features, vertical layered features, and spatial correlation features based on an attention mechanism and a fully connected network to obtain each fused feature vector, and generate the phytoplankton density of each grid area based on each fused feature vector; Among them, the recognition model is obtained by training the initial recognition model based on several historical phytoplankton density data and several historical water temperature data sets in the target ocean area at different periods. Specifically, the historical phytoplankton density data is used as the proportion label, and the historical water temperature data set is used as the data feature. Several label proportion training data packets are constructed and the initial recognition model is trained on the label proportion learning to obtain the recognition model. The initial recognition model is obtained based on the deep learning model.

9. The system for identifying phytoplankton density based on ocean water temperature according to claim 8, wherein: The recognition model performs feature fusion on each of the temporal dynamic features, vertical hierarchical features, and spatial correlation features based on the attention mechanism and the fully connected network to obtain each fused feature vector, and generates the phytoplankton density of each grid area according to each fused feature vector, including: Performing feature encoding on each of the temporal dynamic features and each of the vertical hierarchical features through a preset bidirectional gated recurrent unit to obtain corresponding temporal feature vectors and vertical feature vectors; Performing multi-scale pooling on each of the spatial correlation features through a preset pooling layer to obtain each corresponding spatial feature vector; For each of the grid areas, the temporal feature vector, the vertical feature vector, and the spatial feature vector of the grid area are input into a preset fully connected layer, so that the fully connected layer generates an attention weight corresponding to each feature vector based on an attention mechanism; According to each of the attention weights, the temporal feature vectors, vertical feature vectors, and spatial feature vectors of each of the grid areas are subjected to feature fusion through a fully convolutional network and a LeakyReLU activation function to generate a fused feature vector corresponding to each grid area; According to each of the fused feature vectors, the phytoplankton density of each of the grid areas is generated by using a Softmax function and label ratio learning parameters.

10. A phytoplankton density identification system based on ocean water temperature according to claim 8 or 9, characterized in that: The phytoplankton density identification system further includes a model training module, which is used to train an initial identification model based on a number of historical phytoplankton density data and a number of historical water temperature data sets in a target ocean area at different periods to obtain the identification model, including a model construction unit, a training data set division unit, a feature extraction unit, a training data construction unit, and a training unit; The model construction unit is used to obtain the initial recognition model based on the deep learning model, and the initial recognition model includes an input layer, a bidirectional gated recurrent unit, a pooling layer, a fully connected layer, a convolutional layer and an output layer; The training data set division unit is used to divide the plurality of training phytoplankton density data and the plurality of training water temperature data sets corresponding to each grid area according to the grid area division of the target ocean area, the plurality of historical phytoplankton density data and the plurality of historical water temperature data sets; The feature extraction unit is used to extract features from each of the training water temperature data sets to obtain each corresponding training water temperature feature set; The training data construction unit is used to create a plurality of label ratio training data packets for each grid area according to each of the training phytoplankton density data and each of the training water temperature feature sets, wherein any of the training phytoplankton density data is used as a ratio label in the corresponding label ratio training data packet, and any of the training water temperature feature sets is used as a data feature in the corresponding label ratio training data packet; The training unit is used to train the initial recognition model according to a preset hybrid loss function and each of the label ratio training data packets, determine the deep learning parameters and the label ratio learning parameters, and then update the recognition model, wherein the hybrid loss function includes a deep learning loss function and a label ratio learning loss function, and the label ratio learning loss function is constructed based on minimizing the Wasserstein distance between the prediction result and the proportional label.

Images