A method and system for underwater dissolved oxygen reconstruction based on deep learning

Abstract

The present application relates to a method and system for underwater dissolved oxygen reconstruction based on deep learning. The method uses higher-resolution gridded data as input data, and reconstructs the dissolved oxygen content of multiple layers underwater based on a reconstruction model constructed based on a deep neural network. Compared with the currently used linear statistics and machine learning solutions, the present method uses ocean gridded data with a more complete temporal and spatial coverage scale. On the one hand, the reconstruction model can fully mine the more detailed and accurate spatiotemporal feature information contained in the gridded data. On the other hand, since the gridded data includes sea surface ecological elements, and the reconstruction process only relies on sea surface data, the ecological elements are included in the model input here, achieving a higher-density and more comprehensive reconstruction of the mapping relationship between surface ocean elements and underwater multi-layer dissolved oxygen, which not only solves the problem of lack of underwater data, but also improves the accuracy of underwater dissolved oxygen content reconstruction.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of dissolved oxygen reconstruction technology, and in particular, to a method and system for underwater dissolved oxygen reconstruction based on deep learning. Background Art

[0002] Dissolved oxygen (DO) is not only an important indicator of water quality, but also closely related to the life activities of marine organisms. The formation mechanism of the minimum dissolved oxygen area (hypoxic zone) is the focus of researchers. Regarding dissolved oxygen reconstruction, most existing methods are limited to using scattered data and reconstructing using machine models, but the accuracy of underwater dissolved oxygen reconstruction using this method is low. Summary of the invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] The main purpose of the embodiments of the present disclosure is to propose a method and system for underwater dissolved oxygen reconstruction based on deep learning, which can improve the accuracy of underwater dissolved oxygen content reconstruction.

[0005] A first aspect of an embodiment of the present application proposes an underwater dissolved oxygen reconstruction method based on deep learning, the method comprising:

[0006] Acquire gridded data of the ocean, wherein the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen;

[0007] Construct a reconstruction model of underwater dissolved oxygen based on deep neural network;

[0008] The dissolved oxygen content of underwater multiple layers is reconstructed based on the gridded data through the reconstruction model.

[0009] The underwater dissolved oxygen content reconstruction method based on deep learning provided in this embodiment uses higher-resolution grid data as input data, and reconstructs the dissolved oxygen content of multiple layers underwater based on a reconstruction model constructed based on a deep neural network; compared with the currently used linear statistics and machine learning solutions, this method uses ocean grid data with a more complete temporal and spatial coverage scale. On the one hand, the reconstruction model can fully mine the more detailed and accurate spatiotemporal feature information contained in the grid data; on the other hand, since the grid data includes sea surface ecological elements, and the reconstruction process only relies on sea surface data, the ecological elements are included in the model input here, achieving a higher density and more comprehensive reconstruction of the mapping relationship between surface ocean elements and underwater multi-layer dissolved oxygen, which not only solves the problem of lack of underwater data, but also improves the accuracy of underwater dissolved oxygen content reconstruction.

[0010] In some embodiments of the present application, the reconstruction model includes a convolution module and a fully connected module connected to the convolution module, the convolution module includes at least one basic module, the basic module includes a convolution layer, an attention layer and a maximum pooling layer, the convolution layer is provided with an activation function, and the attention layer is provided with a channel attention mechanism; the fully connected module includes at least one fully connected layer;

[0011] The underwater dissolved oxygen content reconstructed by the reconstruction model and based on the gridded data includes:

[0012] The sea surface ecological elements and the sea surface physical elements are used as channels of the convolution module, the sea surface ecological elements and the sea surface physical elements are input into the convolution module, the initial feature map corresponding to the channel is extracted by the convolution layer, the initial feature map is enhanced by the channel attention mechanism in the attention layer to obtain a first feature map, and the maximum pooling layer performs maximum pooling processing on the first feature map to obtain a second feature map;

[0013] The second feature map is input into the fully connected module to reconstruct the dissolved oxygen content under water through the fully connected layer.

[0014] In some embodiments of the present application, the initial feature map is enhanced by the channel attention mechanism in the attention layer to obtain the first feature map, including:

[0015] Convolving the initial feature map corresponding to the channel to obtain a first intermediate feature map;

[0016] The first intermediate feature map is aggregated into a size of The second intermediate feature map, where the number of channels is ;

[0017] The dependency between the channels is captured from the second intermediate feature map by the following formula to obtain the third intermediate feature map:

[0018]

[0019] in, represents the sigmoid activation function, represents the ReLU activation function, represents the second intermediate feature map, represents the third intermediate feature map, represents the corresponding weight parameter of the first fully connected layer in the attention layer, Represents the corresponding weight parameters of the second fully connected layer;

[0020] The first intermediate feature map and the third intermediate feature map are multiplied in the channel dimension to obtain a first feature map.

[0021] In some embodiments of the present application, obtaining gridded data includes:

[0022] Obtain CMEMS reanalysis gridded datasets;

[0023] The gridded data is obtained by selecting and combining sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen from the reanalysis gridded data set.

[0024] In some embodiments of the present application, the sea surface ecological elements include at least one of the following:

[0025] CO2 surface partial pressure, phytoplankton carbon concentration, chlorophyll concentration, nitrate concentration, phosphate concentration, silicate concentration, dissolved oxygen concentration, dissolved iron concentration and net primary productivity.

[0026] In some embodiments of the present application, the sea surface physical element includes at least one of the following:

[0027] Seawater potential temperature, eastward current velocity, northward current velocity, sea surface height, ocean mixed layer thickness and salinity.

[0028] A second aspect of an embodiment of the present application provides an underwater dissolved oxygen reconstruction system based on deep learning, the system comprising:

[0029] A data acquisition unit, used to acquire gridded data of the ocean, wherein the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen;

[0030] A model building unit, used to build a reconstruction model of underwater dissolved oxygen based on a deep neural network;

[0031] A data reconstruction unit is used to reconstruct the dissolved oxygen content of multiple layers underwater through the reconstruction model and based on the gridded data.

[0032] In some embodiments of the present application, the reconstruction model includes a convolution module and a fully connected module connected to the convolution module, the convolution module includes at least one basic module, the basic module includes a convolution layer, an attention layer and a maximum pooling layer, the convolution layer is provided with an activation function, and the attention layer is provided with a channel attention mechanism; the fully connected module includes at least one fully connected layer;

[0033] The data reconstruction unit comprises:

[0034] A first processing unit is used to use the sea surface ecological elements and the sea surface physical elements as channels of the convolution module, input the sea surface ecological elements and the sea surface physical elements into the convolution module, extract an initial feature map corresponding to the channel through the convolution layer, enhance the initial feature map through the channel attention mechanism in the attention layer to obtain a first feature map, and perform maximum pooling processing on the first feature map through the maximum pooling layer to obtain a second feature map;

[0035] The second processing unit is used to input the second feature map into the fully connected module to reconstruct the underwater dissolved oxygen content through the fully connected layer.

[0036] A third aspect of an embodiment of the present application proposes an electronic device, comprising at least one controller and a memory for communicating with the controller; the memory stores instructions that can be executed by the at least one controller, and the instructions are executed by at least one of the controllers so that at least one of the controllers executes the underwater dissolved oxygen reconstruction method based on deep learning as described above.

[0037] A fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed, the underwater dissolved oxygen reconstruction method based on deep learning as described above is implemented.

[0038] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 It is a flow chart of an embodiment of a method for underwater reconstruction of dissolved oxygen based on deep learning provided by the present application;

[0041] Figure 2 is an architectural diagram of an embodiment of a reconstruction model provided by the present application;

[0042] Figure 3 This application provides Figure 2 An architectural diagram of an embodiment of an attention layer in FIG.

[0043] Figure 4It is an architectural diagram of an embodiment of an electronic device provided by the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] Dissolved oxygen (DO) in seawater is a marker for measuring the hypoxia of water bodies, and is also an important indicator of seawater quality, sea-air interaction, and the growth of marine organisms. It is generally believed that when the O2 (oxygen) concentration in water bodies is lower than 63 micromoles per kilogram (umol / kg), hypoxia will occur, which will make it difficult for most aquatic organisms to maintain their normal life activities, and may even induce biochemical processes, release toxic gases, and deteriorate water quality.

[0046] In the DO reconstruction method, the adaptability of statistical linear models is poor and cannot adapt to nonlinear or complex situations with interference. Existing machine learning algorithms do not take biological factors into account as factors in the model. The application of these algorithms relies on calculating the DO of a certain location underwater through other factor data, which makes the model have considerable spatial limitations, and there is room for improvement in the accuracy of underwater dissolved oxygen reconstruction.

[0047] like Figure 1 In one embodiment of the present application, a method for underwater dissolved oxygen reconstruction based on deep learning is provided, the method comprising the following steps S110 to S130:

[0048] Step S110: Acquire ocean grid data.

[0049] In this step, the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen, or referred to as sea surface ecological variables and sea surface physical variables.

[0050] In some embodiments, step S110 includes:

[0051] Step S111: Obtain a CMEMS reanalysis gridded data set.

[0052] Step S112: Select sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen from the reanalysis gridded data set to obtain gridded data.

[0053] CMEMS refers to Copernicus Marine Environment Monitoring Service. The reanalysis gridded datasets obtained from CMEMS include two reanalysis gridded datasets: Global Ocean Biogeochemical Report Data and Global Ocean Ensemble Physical Reanalysis Data. These two reanalysis gridded datasets provide common biogeochemical and physical element fields, and both datasets add forcing or constraints to the data to make their results as close to the observations as possible, and they have all undergone good quality control.

[0054] It should be noted that this embodiment can also use SODA reanalysis data (https: / / dsrs.atmos.umd.edu / DATA / ) or CMIP6 simulated dissolved oxygen (https: / / esgf-node.llnl.gov / projects / cmip6 / ) to obtain gridded data. This embodiment prefers the CMEMS dataset, which includes two reanalysis grids with more complete temporal and spatial coverage scales, higher resolution and accuracy.

[0055] The temporal resolution of the CMEMS reanalysis gridded dataset includes daily and monthly scales. Monthly average data can be selected here, and all elements in the reanalysis gridded dataset that may be related to dissolved oxygen (i.e., ecological elements and physical elements) are taken into consideration for subsequent use.

[0056] In some embodiments, the elements include fifteen elements, namely, sea surface ecological elements and sea surface physical elements. Among them, the ecological elements include: CO2 (carbon dioxide) surface partial pressure, phytoplankton carbon concentration, chlorophyll concentration, nitrate concentration, phosphate concentration, silicate concentration, dissolved oxygen concentration, dissolved iron concentration, and net primary productivity; among them, the physical elements include: seawater potential temperature, salinity, eastward seawater flow velocity, northward seawater flow velocity, sea surface height, and ocean mixed layer thickness.

[0057] Here, the relationship between dissolved oxygen in different water layers and other selected factors in the surface layer (the fifteen factors mentioned above) was tested. Since the data did not conform to the normal distribution, the Spearman's Rank Correlation Coefficient test was selected, as shown below (1);

[0058] (1)

[0059] in, is the difference in rank between each pair of observations, is the number of observations, here the sample size is 10000.

[0060] The results show that there is no strong correlation between underwater dissolved oxygen and sea surface ecological and physical factors ( >0.5), with the increase of depth, the correlation between the factors and DO did not show a large fluctuation and was relatively stable.

[0061] Step S120: construct a reconstruction model of underwater dissolved oxygen based on a deep neural network.

[0062] In this embodiment, the reconstruction model is constructed based on a deep neural network, and the reconstruction model is used to reconstruct the dissolved oxygen content under the ocean.

[0063] In some embodiments, LeNet can be used to build a dissolved oxygen reconstruction model, or a neural network such as ResNet or VGGNet can be used to build a reconstruction model, which is not specifically limited here. The input data of the reconstruction model is grid data including physical elements and ecological elements, and the output is the dissolved oxygen content at different depths underwater. It should also be noted that this embodiment omits the conventional training process of the model and only introduces the structure of the model and the use process of the model.

[0064] Step S130: reconstruct the underwater dissolved oxygen content by reconstructing the model based on the gridded data.

[0065] In this step, the gridded data is input into the reconstruction model, and the gridded data is used as a channel. The model then performs feature learning on the channel, and finally obtains the underwater dissolved oxygen content output by the model.

[0066] This method uses higher-resolution gridded data as input data, and reconstructs the dissolved oxygen content of multiple underwater layers based on a reconstruction model built based on a deep neural network. Compared with the currently used linear statistics and machine learning solutions, this method uses ocean gridded data with a more complete temporal and spatial coverage scale. On the one hand, the reconstruction model can fully mine the more detailed and accurate spatiotemporal feature information contained in the gridded data. On the other hand, since the gridded data includes sea surface ecological elements and the reconstruction process only relies on sea surface data, the ecological elements are included in the model input here, achieving a higher-density and more comprehensive reconstruction of the mapping relationship between surface ocean elements and underwater multi-layer dissolved oxygen, which not only solves the problem of lack of underwater data, but also improves the accuracy of underwater dissolved oxygen content reconstruction.

[0067] In some embodiments of the present application, the reconstruction model includes a convolution module and a fully connected module connected to the convolution module, the convolution module includes at least one basic module, the basic module includes a convolution layer, an attention layer and a maximum pooling layer, the convolution layer is provided with an activation function, and the attention layer is provided with a channel attention mechanism; the fully connected module includes at least one fully connected layer.

[0068] The dissolved oxygen content of the underwater multiple layers reconstructed by reconstructing the model based on the gridded data in step S130 includes:

[0069] Step S1310: Using sea surface ecological elements and sea surface physical elements as channels of the convolution module, inputting the sea surface ecological elements and sea surface physical elements into the convolution module, extracting the initial feature map corresponding to the channel through the convolution layer, enhancing the initial feature map through the channel attention mechanism in the attention layer to obtain a first feature map, and performing maximum pooling processing on the first feature map through the maximum pooling layer to obtain a second feature map;

[0070] Step S1320: input the second feature map into the fully connected module to reconstruct the underwater dissolved oxygen content through the fully connected layer.

[0071] Among them, ecological elements and physical elements are used as channels. For example, if there are 15 elements, there are 15 channels. Features are extracted through the convolution module, and the extracted features are enhanced through the channel attention mechanism in the convolution module. Finally, the fully connected layer is used as the output layer to reconstruct the dissolved oxygen content of multiple layers underwater. It should be noted that multiple basic modules and multiple fully connected layers can be included here.

[0072] In some embodiments, LeNet is used as a basic model, LeNet includes a convolution module and a fully connected module, the convolution module includes multiple stacked basic modules, the basic module includes a convolution layer and a maximum pooling layer, and each convolution layer includes a corresponding activation function; the fully connected module is connected to the convolution module, that is, the output of the convolution module is used as the input of the fully connected module, and the fully connected module includes multiple fully connected layers. It should be noted that LeNet is a well-known model, and the specific structure is not described here. On the basis of the LeNet basic model, a channel attention mechanism is added to the basic module (that is, an attention layer is added) so that the basic module forms the basic module described in the above embodiment. Because the number of channels is the same as the number of elements, when the number of element inputs is relatively large, the number of channels is also large, then adding a channel attention mechanism can enable the model to globally aggregate the information of each channel, so that the model can better autonomously distinguish the influence of different elements on dissolved oxygen during the training and learning process, optimize the reconstructed model, and ultimately improve the model effect.

[0073] In some embodiments of the present application, in the above step S1310, the initial feature map is enhanced by the channel attention mechanism in the attention layer to obtain the first feature map, including the following:

[0074] Step S1311: Convolve the initial feature map corresponding to the channel to obtain a first intermediate feature map.

[0075] Step S1312: Aggregate the first intermediate feature map into a size of The second intermediate feature map of .

[0076] Step S1313: Capture the dependency between channels from the second intermediate feature map by the following formula (2) to obtain a third intermediate feature map:

[0077] (2)

[0078] in, represents the sigmoid activation function, represents the ReLU activation function, represents the second intermediate feature map, represents the third intermediate feature map, represents the corresponding weight parameter of the first fully connected layer in the attention layer, Represents the corresponding weight parameters of the second fully connected layer.

[0079] Step S1314: multiply the first intermediate feature map and the third intermediate feature map in the channel dimension to obtain a first feature map.

[0080] In this embodiment, assuming that there are 15 elements, there are 15 channels. First, multiple initial feature maps corresponding to multiple channels are convolved to obtain multiple corresponding first intermediate feature maps. Then, the information of multiple channels is aggregated globally through global average pooling, and the multiple first intermediate feature maps are aggregated into a size of The second intermediate feature map is then adaptively generated through a formula to generate the importance weight of each channel. The formula includes two fully connected layers to parameterize the channel selection mechanism, that is, to fuse the feature maps of each channel. There are two activation functions for activation in the public display. The s in the formula is used to characterize the multiple weight values ​​corresponding to multiple channels. Finally, multiplication in the channel dimension is equivalent to multiplying the initial feature map corresponding to each channel by the corresponding weight value, re-weighting and adjusting the initial feature map, and finally completing the enhancement of the initial feature map.

[0081] The method provided in this embodiment globally aggregates the information of each channel, uses global average pooling to transform the initial feature map into a feature vector map of length C, and then adaptively generates the importance weight of each channel, and re-weights the initial feature map by multiplying the final channel dimension. This embodiment can effectively enhance the feature weights represented by important channels, while suppressing the features of unimportant channels, that is, reducing the influence of unimportant or irrelevant elements, thereby achieving the effect of optimizing the model.

[0082] In order to facilitate the understanding of those skilled in the art, an embodiment is provided, taking the equatorial Pacific as the research area (spatial range 30S-23N, 156E-85W), a method for underwater dissolved oxygen reconstruction based on deep learning, comprising:

[0083] Step S910: Select multiple marine ecological elements and marine physical elements.

[0084] Monthly average data were selected from CMEMS, including fifteen variable elements: surface partial pressure of CO2, phytoplankton carbon concentration, chlorophyll concentration, nitrate concentration, phosphate concentration, silicate concentration, dissolved oxygen concentration, dissolved iron concentration, net primary productivity, seawater potential temperature, salinity, eastward seawater flow velocity, northward seawater flow velocity, sea surface height, and ocean mixed layer thickness.

[0085] Step S920: Set the reconstruction model.

[0086] Since oxygen loss in equatorial waters is most severe in the intermediate waters (100-600 meters), the largest absolute and relative losses occur in the main thermocline at depths of 100-300 meters and in the mid-depth water masses (1,000-2,000 meters). The key 21 layers depths=[50,100,120,150,180,200,220,240,270,300,330,370,400,450,500,600,800,1000,1200,1500,2000] were selected.

[0087] Mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and mean bias (MB) were used as evaluation criteria for model selection.

[0088] (3)

[0089] (4)

[0090] (5)

[0091] (6)

[0092] in, represents the true value of the ith observation, represents the predicted value of the ith observation, and n represents the total number of observations.

[0093] The above are all commonly used error evaluation indicators in statistical analysis. MSE and RMSE measure the average of the squared errors between the predicted value and the actual value, MAE represents the absolute difference, and MB directly displays the actual error.

[0094] like Figure 2 As shown, the embodiment makes changes to the traditional LeNet network. In addition to adding pooling layers, convolutional layers, and fully connected layers, channel attention is also added, which allows the model to make full use of the channel attention mechanism to enhance feature representation capabilities.

[0095] The 15 selected elements are used as channels of the model for training, and the channel attention mechanism is introduced into the convolutional network, so that the network can adaptively readjust the importance of each channel. In this way, the model can better and autonomously distinguish the impact of different elements on DO during the training and learning process, thereby obtaining more accurate model reconstruction results.

[0096] like Figure 3 , the operation of the channel attention mechanism is as follows:

[0097] First, multiple initial feature maps corresponding to multiple channels are convolved to obtain multiple corresponding first intermediate feature maps. Then, global average pooling (i.e. Figure 3 Global Pooling in the global aggregation of multiple channels of information, multiple first intermediate feature maps are aggregated into a size of The second intermediate feature map is then adaptively generated through a formula to generate the importance weight of each channel. The formula includes two fully connected layers to parameterize the channel selection mechanism, that is, to fuse the feature maps of each channel. There are two activation functions for activation in the public display. The s in the formula is used to characterize the multiple weight values ​​corresponding to multiple channels. Finally, multiplication in the channel dimension is equivalent to multiplying the initial feature map corresponding to each channel by the corresponding weight value, re-weighting and adjusting the initial feature map, and finally completing the enhancement of the initial feature map.

[0098] The channel attention mechanism set up can effectively enhance the feature weights represented by important channels, while suppressing the features of unimportant channels, that is, reducing the influence of unimportant or irrelevant factors, thereby achieving the effect of optimizing the model.

[0099] Step S930: Use the reconstruction model to reconstruct the multi-layer dissolved oxygen content under the ocean.

[0100] The following provides experimental examples:

[0101] For the control group, for the linear regression model and the integrated machine learning model, the data were divided into trian (training set) and test (test set) in chronological order with a ratio of 8:2;

[0102] In the embodiment of the present application, for the reconstruction model based on deep learning, the data is divided into train, val (validation set) and test in a ratio of 8:1:1, where the increase of val is used to design early stopping, which can effectively prevent overfitting. To ensure the repeatability of the experiment, all random seeds are set to 37.

[0103] (1) Linear regression model;

[0104] Since the input of the statistical linear model must be two-dimensional, in this part, X needs to be flattened into a shape of (n, 15) and Y needs to be flattened into a shape of (n, 21), where 15 is the number of variables and 21 is the number of reconstruction DO layers. In order to eliminate the impact of the dimension on the results and improve the stability of the linear regression model, the data is standardized before fitting the model.

[0105] Using the stepwise regression method, the prediction factors were gradually introduced or removed, and the best regression model factor combination ['Sal', 'PO4', 'Temp', 'Si', 'NO3', 'spCO2', 'vo', 'MLD', 'uo', 'Fe', 'PhyC', 'NPP'] was selected, with a total of 11 factors. Then the multicollinearity VIF test was performed to ensure that there was no strong correlation between the independent variables. The VIF of all variables was <5, indicating that although the selected variables had a certain degree of multicollinearity, it was acceptable.

[0106] Since DO is a continuous dependent variable, the ordinary least squares method OLS is selected to fit the linear regression model. The fitting results are shown in Table 1 below:

[0107] Table 1

[0108]

[0109] (2) Machine learning models;

[0110] In this part, ERT and RF algorithms were selected. Since this experiment also requires two-dimensional input and the amount of data is large (tens of millions), after considering (balancing) the efficiency of computing resources and the effect of the machine learning model, the hyperparameters are set as follows: n_estimators=100, max_depth=30,min_samples_leaf=50, max_features='sqrt'. In this embodiment, all elements are first input into the machine learning model for fitting, and then eight elements are selected according to the feature importance ranking for refitting, and the new machine learning model is obtained before the result is calculated.

[0111] Compare the model effects before and after feature selection. All indicators are the average of the reconstructed 21-layer DO. ERT and RF can learn better weights of feature variables. If computing power allows, as comprehensive factors as possible can be input, as shown in Table 2.

[0112] Table 2

[0113]

[0114] (3) Reconstruction model (based on LeNet with channel attention mechanism);

[0115] The processed variable X data shape is (360, 15, 213, 477), and the target Y data shape is (360, 21, 213, 477), where 360 ​​is the month covered by the data, 15 is the number of variables, and 21 is the number of DO layers to be output. (213, 477) represents the location information.

[0116] The loss function of the model is designed to be MSE, the learning rate is set to 0.001, the optimizer is set to Adam, and all variables X are normalized. The formula is as follows:

[0117] (7)

[0118] In terms of land processing, a land mask is designed so that the land output results of each internal cycle of the reconstruction model are assigned to zero, so that all the information learned by the model in the land part cannot be transmitted back to the reconstruction model weights, thereby eliminating the impact of deleting land on the learning effect of the reconstruction model.

[0119] The results of Table 3 show that the reconstructed model with attention mechanism has better performance than LeNet and converges faster. The results of the reconstructed model on the test set are MAE=7.98 and RMSE=12.00.

[0120] Table 3

[0121]

[0122] Eight layers of the reconstruction results of a certain month (May 2020) were randomly selected for drawing demonstration. According to the experimental results, the reconstruction results of the reconstruction model can basically reproduce the distribution of the real values.

[0123] According to the deviation diagram between the reconstruction results of LeNet and the reconstruction model and the true value (not shown in the figure), the reconstruction model is better than LeNet in learning details, so that the area of ​​​​the deviation is reduced.

[0124] In addition, the mean DO value (120.5759 μmol / kg) learned by the reconstruction model is closer to the true value than LeNet (119.1019 μmol / kg) and ResNet (116.3777 μmol / kg).

[0125] An embodiment is also provided, an underwater dissolved oxygen reconstruction system based on deep learning, the system comprising:

[0126] The data acquisition unit is used to acquire gridded data of the ocean, wherein the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen.

[0127] The model building unit is used to build a reconstruction model of underwater dissolved oxygen based on deep neural network.

[0128] The data reconstruction unit is used to reconstruct the dissolved oxygen content of underwater multiple layers based on the gridded data by reconstructing the model.

[0129] It should be noted that the present system embodiment and the above-mentioned method embodiment are based on the same inventive concept, so the relevant contents of the above-mentioned method embodiment are also applicable to the present system embodiment and will not be repeated here.

[0130] Reference Figure 4 The embodiment of the present application further provides an electronic device, the electronic device comprising:

[0131] at least one memory;

[0132] at least one processor;

[0133] at least one program;

[0134] The program is stored in the memory, and the processor executes at least one program to implement the underwater dissolved oxygen reconstruction method based on deep learning described above in the present disclosure.

[0135] The electronic device may be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a vehicle-mounted computer, etc.

[0136] The electronic device according to the embodiment of the present application is described in detail below.

[0137] The processor 1600 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0138] The memory 1700 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1700 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the program code is stored in the memory 1700, and the processor 1600 calls and executes the underwater dissolved oxygen reconstruction method based on deep learning in the embodiment of the present application.

[0139] Input / output interface 1800, used to implement information input and output;

[0140] Communication interface 1900, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0141] Bus 2000 , which transmits information between various components of the device (e.g., processor 1600 , memory 1700 , input / output interface 1800 , and communication interface 1900 );

[0142] The processor 1600 , the memory 1700 , the input / output interface 1800 , and the communication interface 1900 are connected to each other in communication within the device via the bus 2000 .

[0143] An embodiment of the present application also provides a storage medium, which is a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the above-mentioned underwater dissolved oxygen reconstruction method based on deep learning.

[0144] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0145] The embodiments described in this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0146] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0147] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0149] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0150] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0151] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0153] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0154] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store programs.

[0155] The above is a specific description of the preferred implementation of the embodiments of the present application, but the embodiments of the present application are not limited to the above-mentioned implementation methods. Technical personnel familiar with the field can also make various equivalent modifications or substitutions without violating the spirit of the embodiments of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the embodiments of the present application.

Claims

1. A method for underwater dissolved oxygen reconstruction based on deep learning, characterized in that: The method comprises: Acquire gridded data of the ocean, wherein the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen; Constructing a reconstruction model of underwater dissolved oxygen based on a deep neural network; the reconstruction model includes a convolution module and a fully connected module connected to the convolution module, the convolution module includes at least one basic module, the basic module includes a convolution layer, an attention layer and a maximum pooling layer, the convolution layer is provided with an activation function, and the attention layer is provided with a channel attention mechanism; the fully connected module includes at least one fully connected layer; Reconstructing the dissolved oxygen content of multiple layers underwater based on the gridded data by using the reconstruction model includes: The sea surface ecological elements and the sea surface physical elements are used as channels of the convolution module, the sea surface ecological elements and the sea surface physical elements are input into the convolution module, the initial feature map corresponding to the channel is extracted by the convolution layer, the initial feature map is enhanced by the channel attention mechanism in the attention layer to obtain a first feature map, and the maximum pooling layer performs maximum pooling processing on the first feature map to obtain a second feature map; Inputting the second feature map into the fully connected module to reconstruct the dissolved oxygen content under water through the fully connected layer; The dissolved oxygen content of underwater multiple layers is reconstructed based on the gridded data through the reconstruction model.

2. The underwater dissolved oxygen reconstruction method based on deep learning according to claim 1 is characterized in that: The step of enhancing the initial feature map by the channel attention mechanism in the attention layer to obtain a first feature map includes: Convolving the initial feature map corresponding to the channel to obtain a first intermediate feature map; The first intermediate feature map is aggregated into a size of The second intermediate feature map, where the number of channels is ; The dependency between the channels is captured from the second intermediate feature map by the following formula to obtain the third intermediate feature map: ； in, represents the sigmoid activation function, represents the ReLU activation function, represents the second intermediate feature map, represents the third intermediate feature map, represents the corresponding weight parameter of the first fully connected layer in the attention layer, Represents the corresponding weight parameters of the second fully connected layer; The first intermediate feature map and the third intermediate feature map are multiplied in the channel dimension to obtain a first feature map.

3. The underwater dissolved oxygen reconstruction method based on deep learning according to claim 1 is characterized in that: The obtaining of ocean grid data includes: Obtain CMEMS reanalysis gridded datasets; The gridded data is obtained by selecting and combining sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen from the reanalysis gridded data set.

4. The underwater dissolved oxygen reconstruction method based on deep learning according to claim 1 is characterized in that: The sea surface ecological factors include at least one of the following: CO2 surface partial pressure, phytoplankton carbon concentration, chlorophyll concentration, nitrate concentration, phosphate concentration, silicate concentration, dissolved oxygen concentration, dissolved iron concentration and net primary productivity.

5. The underwater dissolved oxygen reconstruction method based on deep learning according to claim 1 is characterized in that: The sea surface physical elements include at least one of the following: Seawater potential temperature, eastward current velocity, northward current velocity, sea surface height, ocean mixed layer thickness and salinity.

6. An underwater dissolved oxygen reconstruction system based on deep learning, characterized in that: The system comprises: A data acquisition unit, used to acquire gridded data of the ocean, wherein the gridded data includes sea surface ecological elements and sea surface physical elements associated with underwater dissolved oxygen; A model building unit, used to build a reconstruction model of underwater dissolved oxygen based on a deep neural network; A data reconstruction unit, used to reconstruct the dissolved oxygen content of multiple layers underwater based on the gridded data through the reconstruction model; the reconstruction model includes a convolution module and a fully connected module connected to the convolution module, the convolution module includes at least one basic module, the basic module includes a convolution layer, an attention layer and a maximum pooling layer, the convolution layer is provided with an activation function, and the attention layer is provided with a channel attention mechanism; the fully connected module includes at least one fully connected layer; The data reconstruction unit comprises: A first processing unit is used to use the sea surface ecological elements and the sea surface physical elements as channels of the convolution module, input the sea surface ecological elements and the sea surface physical elements into the convolution module, extract an initial feature map corresponding to the channel through the convolution layer, enhance the initial feature map through the channel attention mechanism in the attention layer to obtain a first feature map, and perform maximum pooling processing on the first feature map through the maximum pooling layer to obtain a second feature map; The second processing unit is used to input the second feature map into the fully connected module to reconstruct the underwater dissolved oxygen content through the fully connected layer.

7. An electronic device, characterized in that: It includes at least one controller and a memory for communicating with the controller; the memory stores instructions that can be executed by the at least one controller, and the instructions are executed by at least one of the controllers to enable at least one of the controllers to perform the underwater dissolved oxygen reconstruction method based on deep learning as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the underwater dissolved oxygen reconstruction method based on deep learning as described in any one of claims 1 to 5.

Images