Ocean surface salinity inversion method, system and equipment and storage medium

Abstract

The invention belongs to the field of ocean remote sensing detection, and particularly relates to an ocean surface salinity inversion method, system and device and a storage medium, and the method comprises the steps: obtaining training day brightness temperature data and training month salinity data of an ocean surface; preprocessing the brightness temperature data of the training day and then carrying out regional screening to obtain regional brightness temperature data; performing interpolation processing on the training month salinity data to obtain interpolation salinity data; performing time alignment on the regional brightness data and the interpolation salinity data to obtain a brightness temperature data set and a salinity data set; extracting features according to the brightness temperature data set to obtain brightness temperature features, inputting the brightness temperature features into the salinity intelligent inversion network model to obtain an output salinity value, calculating the output salinity value and an output loss value of the salinity data set through a loss function, and adjusting model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimization model; and inputting the actual brightness temperature data into the optimization model to obtain a salinity value. The method has the effect of improving the inversion precision.

Description

Technical Field

[0001] The present invention belongs to the field of ocean remote sensing detection, and particularly relates to a method, system, device and storage medium for salinity inversion of the ocean surface. Background Art

[0002] Ocean surface salinity is an important indicator for describing the ocean state, simulating ocean circulation and detecting climate change, and is of great significance for ocean research. In the current context of the continuous progress of remote sensing technology, satellite remote sensing has become an important means to obtain ocean surface salinity data. After obtaining the original brightness temperature data through the satellite, it is necessary to preprocess the original brightness temperature data and use the ocean surface salinity inversion model to convert the brightness temperature data into the corresponding salinity data.

[0003] In related technologies, common ocean surface salinity inversion algorithm models include empirical algorithm models, multiple linear regression models, ridge regression models, etc.

[0004] For the above-mentioned related technologies, the parameters of the empirical algorithm model are usually calibrated for specific regions or scenarios, and in other regions, such as different ocean circulation regions, the generalization ability is poor. The actual salinity of the multiple linear regression model is dominated by non-linear processes (such as salinity fronts, vortex mixing), and the linear model cannot capture complex dynamics. The regularization effect of the ridge regression model depends on sufficient historical data, and the improvement effect is limited when the data is scarce. The above models all have certain limitations, resulting in low accuracy of salinity inversion. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, system, device and storage medium for salinity inversion of the ocean surface, which enhance the expression ability of data features by mapping the extracted feature output to a higher dimension, obtain more information at different scales, and improve the inversion accuracy.

[0006] A method for salinity inversion of the ocean surface includes: Obtaining the training daily brightness temperature data and training monthly salinity data of the ocean surface; Performing preprocessing on the training daily brightness temperature data and then performing regional screening to obtain regional brightness temperature data; Performing interpolation processing on the training monthly salinity data to obtain interpolated salinity data; Aligning the regional brightness data and the interpolated salinity data in time to obtain a brightness temperature data set and a salinity data set; Extracting features from the brightness temperature data set to obtain brightness temperature features, inputting the brightness temperature features into a salinity intelligent inversion network model to obtain an output salinity value, calculating the output loss value between the output salinity value and the salinity data set through a loss function, and adjusting the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; Input the actual brightness temperature data into the optimization model to obtain the salinity value.

[0007] Optionally, the interpolation processing of the training monthly salinity data to obtain the interpolated salinity data includes: According to the training monthly salinity data, obtain the target coordinate values, the salinity corresponding to the target coordinate values, the four coordinate values closest to the target coordinate, and the salinity corresponding to the four coordinate values; Substitute the target coordinate values, the salinity corresponding to the target coordinate values, the four coordinate values closest to the target coordinate, and the salinity corresponding to the four coordinate values into the interpolation formula to obtain the interpolated salinity data; The interpolation formula is: where x and y are the horizontal and vertical coordinates of the target coordinate values, , , and are the four coordinate values around the target coordinate, is the interpolated salinity data, and S is the salinity corresponding to the four coordinate values.

[0008] Optionally, the preprocessing of the training daily brightness temperature data and then regional screening to obtain the regional brightness temperature data includes: Convert the format of the training daily brightness temperature data to obtain the converted data; Correct the converted data to obtain the corrected data; Set the flag bits, where the flag bits include the ocean flag bit and other flag bits. The ocean flag bit is 1 and the other flag bits are 0; Multiply the corrected data and the flag bits to obtain the flag bit data; Obtain the screening area; Screen the flag bit data according to the screening area to obtain the regional brightness temperature data.

[0009] Optionally, extracting features from the brightness temperature dataset to obtain the brightness temperature features includes: Perform time series partitioning on the brightness temperature dataset to obtain a number of patch data, perform one-dimensional convolution operations on the patch data to obtain the convolution data, and the brightness temperature dataset is expressed as: H×W×C; where H represents the number of grid points for constructing the brightness temperature dataset, W represents the time length of the constructed brightness temperature dataset, which is 30 days; C represents the number of channels of the brightness temperature dataset; Map the convolution data to a fixed embedding dimension D through a linear projection layer to obtain the low-dimensional brightness temperature feature data; Input the low-dimensional brightness temperature feature data into the fully connected layer, self-attention layer, position encoding layer, and normalization layer in sequence, and use the high-dimensional brightness temperature data as the brightness temperature feature; The fully connected layer is expressed as: ; Among them, X is the low-dimensional brightness temperature data output by the patch embedding layer and the position encoding layer, with a shape of [H, W / P, D], and Z is the high-dimensional brightness temperature data output after passing through the fully connected layer, with a shape of [H, W / P, M], is a non-linear activation function, b is the learnable weight bias of the fully connected layer, is the learnable weight matrix of the fully connected layer, and the matrix shape is [D, M], is the dynamic gating factor based on the ocean physics prior, and M is the dimension of the high-dimensional brightness temperature data.

[0010] Optionally, the method for obtaining the weight matrix is: Obtain the mean, variance, maximum brightness temperature value of each month, and minimum brightness temperature value of each month of the low-dimensional brightness temperature feature data; Substitute the mean, variance, maximum brightness temperature value of each month, and minimum brightness temperature value of each month into the weight matrix formula to obtain the weight matrix; The weight matrix formula is: ; Among them, f is the mapping function, is the mean, is the variance, max is the maximum brightness temperature value of each month, min is the minimum brightness temperature value of each month, is the weight matrix.

[0011] Optionally, obtaining the dynamic gating factor includes: Obtain the mean brightness temperature of the ocean surface in the current month; Obtain the average brightness temperature of the patch data; According to the mean brightness temperature, the average brightness temperature, and the judgment formula, obtain the dynamic gating factor; The judgment formula is: ; Among them, H is the number of grid points, W is the time length of the constructed brightness temperature data set, C represents the number of channels of the brightness temperature data set, P is the segmentation length, represents the average brightness temperature value of the original brightness temperature observation value in the time dimension within the W / P-th time patch at grid point H, is the average brightness temperature, and are learning parameters, The value range of is 1.5 to 2.0, The value range of

[0012] Optionally, the loss function is expressed as: ; where n represents the number of grid points in the target ocean area, represents the measured salinity value corresponding to the i-th grid point, represents the salinity value inverted by the inversion model corresponding to the i-th grid point of the model, and MAE is the value of the loss function.

[0013] A salinity inversion system for the ocean surface, comprising: An acquisition module, configured to acquire the training daily brightness temperature data and training monthly salinity data of the ocean surface; A preprocessing module, configured to perform preprocessing on the training daily brightness temperature data and then perform region screening to obtain region brightness temperature data; Perform interpolation processing on the training monthly salinity data to obtain interpolated salinity data; An interpolation module, configured to align the region brightness data and the interpolated salinity data in time to obtain a brightness temperature data set and a salinity data set; An optimization module, configured to extract features from the brightness temperature data set to obtain brightness temperature features, input the brightness temperature features into a salinity intelligent inversion network model to obtain an output salinity value, calculate the output loss value of the output salinity value and the salinity data set through a loss function, and adjust the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; An output module, configured to input actual brightness temperature data into the optimized model to obtain a salinity value.

[0014] A terminal device, comprising a memory and a processor, where the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, a salinity inversion method for the ocean surface is adopted.

[0015] A computer-readable storage medium stores a computer program, and when the computer program is loaded and executed by a processor, a salinity inversion method for the ocean surface is adopted.

[0016] The beneficial effects of the present invention are: 1. Preprocess the training day brightness temperature data and then perform regional screening to obtain regional brightness temperature data. Interpolate the training monthly salinity data to obtain interpolated salinity data. Align the regional brightness data and the interpolated salinity data in time to obtain a brightness temperature dataset and a salinity dataset. Extract features from the brightness temperature dataset to obtain brightness temperature features. Input the brightness temperature features into a salinity intelligent inversion network model to obtain an output salinity value. Calculate the output loss value between the output salinity value and the salinity dataset through a loss function. Adjust the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model. Input the actual brightness temperature data into the optimized model to obtain a salinity value. Compared with traditional inversion models, this application inputs brightness temperature features of different scales and dimensions into the inversion network model for training to obtain an optimized model, improving the accuracy of inversion.

[0017] 2. An improved fully connected layer is innovatively added in front of the encoding layer, and a multi-scale convolutional layer is added behind the encoding layer. Compared with the traditional encoder module, adding an improved fully connected layer in front of the encoder can adjust the dimension of the input features and map the input brightness temperature features of low dimension to higher feature dimensions. The input brightness temperature data features after patch embedding and position encoding are relatively simple. Introducing non-linearity through an activation function (ReLU) helps to extract more complex brightness temperature data feature representations in the subsequent encoding layer part, thus improving the inversion accuracy of the model.

[0018] 3. When improving the fully connected layer, the weight matrix is dynamically generated by mapping the spatio-temporal statistical features of the brightness temperature data through a mapping function, with stronger generalization feature extraction ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the image of the brightness temperature data of the present invention before being processed by the flag bit; Figure 2 It is a schematic diagram of the image of the brightness temperature data of the present invention after being processed by the flag bit; Figure 3 It is a schematic diagram of the result after the present invention selects regions for the ocean brightness temperature data; Figure 4 It is a schematic diagram of the model structure of the salinity intelligent inversion network model of the present invention; Figure 5 It is a salinity distribution map of August 2022 obtained by the ocean surface inversion model of the salinity intelligent inversion network model of the present invention; Figure 6 This invention is a salinity distribution map of July 2022 obtained by the ocean surface inversion model of the salinity intelligent inversion network model; Figure 7 It is a schematic diagram of the inversion result obtained by the present invention using the optimized model. DETAILED DESCRIPTION OF THE INVENTION

[0020] A method for retrieving the salinity of the ocean surface, comprising: S1. Obtain the training daily brightness temperature data and training monthly salinity data of the ocean surface; Specifically, obtain the brightness temperature data observed by the satellite daily and the sea surface salinity data measured actually monthly, and process and construct a brightness temperature - salinity data set corresponding to the ocean scenario. A subset of this data set reflects the monthly salinity value and daily brightness temperature value corresponding to each grid point (the center point of the grid unit divided by longitude and latitude on the earth's surface) in a certain year and month of the target ocean area. Specifically, for a subset of the brightness temperature - salinity data set of an ocean scenario, assuming the number of grid points is N, the subset of the data set can be specifically characterized as N rows and 31 columns. The first column is the monthly salinity value corresponding to each grid point, which serves as the training monthly salinity data to be retrieved by the inversion model; the 2nd to 31st columns are the brightness temperature values of each grid point corresponding to that month, which serve as the brightness temperature data feature set of the inversion model.

[0021] S2. Pre - process the training daily brightness temperature data and then perform regional screening to obtain regional brightness temperature data; Pre - processing the training daily brightness temperature data and then performing regional screening to obtain regional brightness temperature data includes: S21. Convert the format of the training daily brightness temperature data to obtain converted data; Specifically, the formats of the training daily brightness temperature data and salinity data are different. The brightness temperature data is in the SMOS satellite data format, and the salinity data is in the Argo format. Convert the SMOS brightness temperature data format to the same NetCDF (.nc) format as the Argo data format.

[0022] S22. Calibrate the converted data to obtain calibrated data; Specifically, perform operations such as geometric calibration, radiometric calibration, and error calibration on the SMOS brightness temperature data to ensure the quality and accuracy of the data.

[0023] S23. Set flag bits, where the flag bits include an ocean flag bit and other flag bits. The ocean flag bit is 1, and other flag bits are 0; S24. Multiply the calibrated data by the flag bits to obtain flag - bit data; Specifically, the ocean flag bit in the land - sea mask of the SMOS data is 1. By setting all flag - bit data other than 1 in the flag - bit data to 0. Then multiply the land - sea mask by the SMOS brightness temperature data to screen out the ocean data in the SMOS original data. Figure 1 and Figure 2 respectively show the specific images of the SMOS original data before and after being processed by the flag bits.

[0024] S25. Obtain the screening area; S26. Screen the flag bit data according to the screening area to obtain the regional brightness temperature data.

[0025] Specifically, the screening area is the area where salinity needs to be obtained. In this actual example, the South China Sea area is selected, with specific longitude and latitude ranging from 110° to 118° east longitude and 8° to 22° north latitude. The data is screened by determining the longitude and latitude, and the corresponding brightness temperature, longitude, and latitude data are extracted. Figure 3 The result after selecting the South China Sea area for the ocean brightness temperature data is shown.

[0026] S3. Perform interpolation processing on the training monthly salinity data to obtain interpolated salinity data; Specifically, perform preprocessing of regional selection and interpolation on the actually measured salinity data to unify the spatio-temporal dimensions of the salinity data and the brightness temperature data, facilitating subsequent data matching. The regional selection processing of Argo data is the same as that of SMOS data, with specific longitude and latitude ranging from 110° to 118° east longitude and 8° to 22° north latitude. The interpolation processing of Argo data can interpolate the accuracy of Argo data from 1°×1° to 0.1°×0.1°, facilitating subsequent matching with SMOS data.

[0027] Performing interpolation processing on the training monthly salinity data to obtain interpolated salinity data includes: According to the training monthly salinity data, obtain the target coordinate values, the salinity corresponding to the target coordinate values, the four coordinate values closest to the target coordinate, and the salinity corresponding to the four coordinate values; Substitute the target coordinate values, the salinity corresponding to the target coordinate values, the four coordinate values closest to the target coordinate, and the salinity corresponding to the four coordinate values into the interpolation formula to obtain the interpolated salinity data; The interpolation formula is: where x and y are the horizontal and vertical coordinates of the target coordinate values, 、 、 and are the four coordinate values around the target coordinate, is the interpolated salinity data, S is the salinity corresponding to the four coordinate values. For example, is the salinity at the coordinate .

[0028] S4. Align the regional brightness data and the interpolated salinity data in time to obtain a brightness temperature data set and a salinity data set; Specifically, the processed satellite brightness temperature data and salinity data for the same month are effectively aligned in terms of data time, and their longitudes and latitudes are matched. Combining the matched data, outliers and incorrect data are removed to ensure the accuracy and consistency of the data, forming a complete sea surface brightness temperature-salinity dataset.

[0029] Perform normalization processing on each subset of the complete sea surface brightness temperature-salinity dataset. Specifically, perform dynamic range normalization on the salinity values to adapt to the salinity differences in different sea areas, and perform 0-255 normalization on the brightness temperature data to simulate the quantization process of remote sensing data.

[0030] Divide the dataset, and divide it into a training set and a test set according to a ratio of 8:2 to obtain the training set and test set required for the inversion model. Specifically, the shape of the training set is [number of grid points, 31]. The first column is the monthly salinity value corresponding to the grid point, and the following 30 columns are the daily brightness temperature values corresponding to the grid point.

[0031] S5. Extract features from the brightness temperature dataset to obtain brightness temperature features, input the brightness temperature features into the salinity intelligent inversion network model to obtain the output salinity value, calculate the output loss value between the output salinity value and the salinity dataset through the loss function, and adjust the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; Extract features from the brightness temperature dataset, and the obtained brightness temperature features include: S51. Divide the brightness temperature dataset into time series to obtain a number of patch data, perform one-dimensional convolution operations on the patch data to obtain convolution data, and the brightness temperature dataset is represented as: H×W×C; Among them, H represents the number of grid points constructing the brightness temperature dataset, W represents the time length of the constructed brightness temperature dataset, which is 30 days; C represents the number of channels of the brightness temperature dataset; S52. Map the convolution data to a fixed embedding dimension D through a linear projection layer and perform position encoding to obtain low-dimensional brightness temperature feature data; S53. Input the low-dimensional brightness temperature feature data into a fully connected layer, a self-attention layer, a position encoding layer, and a normalization layer in sequence, and use the high-dimensional brightness temperature data as the brightness temperature feature; The fully connected layer is represented as: ; Among them, X is the low-dimensional brightness temperature data output by the patch embedding layer and the position encoding layer, with a shape of [H,W / P,D], Z is the high-dimensional brightness temperature data output after passing through the fully connected layer, with a shape of [H,W / P,M], is a non-linear activation function, b is the learnable weight bias of the fully connected layer, is the learnable weight matrix of the fully connected layer, and the matrix shape is [D,M], is a dynamic gating factor based on prior marine physics, where M is the dimension of high-dimensional brightness temperature data.

[0032] Specifically, to fully exploit the local feature information of brightness temperature data at different scales, a multi-scale convolution structure is introduced after the encoder module to improve the accuracy of the inversion results.

[0033] Specifically, the training set of the dataset is input into the salinity intelligent inversion network model. The brightness temperature data of the training set is used as the input feature set, with a shape of [number of grid points, 30], and the salinity data is used as the target set to train the salinity intelligent inversion network model. According to the loss function between the salinity value output by the network and the salinity value of the target set. The gradient of the loss function with respect to the model parameters is calculated through the backpropagation algorithm, and the model parameters are updated using an optimization algorithm according to the gradient of the model parameters. When the loss function is minimized, the training of the model is stopped, and the salinity intelligent inversion network model at this time is determined to be the appropriate ocean surface salinity inversion model.

[0034] The structure of the salinity intelligent inversion network model is as Figure 4 shown. It combines the characteristics of brightness temperature data to optimize and improve the encoding layer. An improved fully connected layer is innovatively added in front of the encoding layer, and a multi-scale convolutional layer is added behind the encoding layer. Compared with the traditional encoder module, adding an improved fully connected layer in front of the encoder can adjust the dimension of the input features, mapping the low-dimensional brightness temperature feature input to a higher feature dimension. After patch embedding and position encoding, the input brightness temperature data features are relatively simple. Introducing non-linearity through the activation function (ReLU) helps extract more complex brightness temperature data feature representations in the subsequent encoding layer part, thus improving the inversion accuracy of the model. Adding a multi-scale convolutional layer behind the encoding layer can capture the detailed changes of brightness temperature data at different time scales and spatial neighborhoods. The traditional encoder module relies on the self-attention mechanism, focusing on capturing global dependence features and lacking the capture of local features of brightness temperature data. The multi-scale convolutional layer makes up for the deficiency of the encoder's ability to perceive local features, thus improving the accuracy of sea surface salinity inversion. In addition, the salinity intelligent inversion network model does not use the decoder in the traditional network for inversion, but achieves a good inversion effect by adjusting the output layer in combination with the characteristics of sea surface salinity inversion.

[0035] Specifically, the input brightness temperature dataset is a tensor of H×W×C, where H represents the number of grid points for constructing the brightness temperature dataset; W represents the time length of the constructed brightness temperature dataset, which is 30 days; C represents the number of channels in the brightness temperature dataset, and the number of channels is taken as C = 1. The patch embedding layer divides the width W of the long-term brightness temperature dataset with patches of length P, and the sequence is divided into H×W / P patches, each patch having a size of 1×P. After flattening the patches, a one-dimensional vector is obtained, which is mapped to a fixed embedding dimension D through a linear projection layer. All patch embeddings form a shape of [H, W / P, D]. After position encoding, the shape of the dataset remains unchanged, and a low-dimensional brightness temperature feature dataset with a shape of [H, W / P, D] is obtained. The optimized and improved encoding layer and the adjusted output layer of this model are described as follows: The optimized and improved encoder module consists of multiple modules. Compared with the traditional encoder module (each module includes a self-attention layer, a multi-layer perceptron, layer normalization, and stochastic depth), there are two improvements. The first is to add an improved fully connected layer in front of the encoder module, which is a fully connected layer that combines physical prior laws and spatio-temporal statistical features of brightness temperature; the second is to add a convolutional layer behind the encoder. These modules are responsible for extracting complex information from the embedded features to obtain high-dimensional features of the brightness temperature dataset.

[0036] Improved fully connected layer: Maps the output of the low-dimensional brightness temperature data after simple feature extraction from the patch embedding layer and the position encoding layer to a higher dimension to enhance the expression ability of data features.

[0037] Self-attention layer: The basic operation of the self-attention mechanism is to map the input features (or patches) to the query (Q), key (K), and value (V) spaces, and aggregate these values by calculating the attention weights to obtain an output that fuses the relevant information between different positions in the input high-dimensional brightness temperature data; then add a residual connection and normalization to this output to restore it to the initial size [H, W / P, M]. Among them, Q, K, and V are obtained by linearly transforming the input X.

[0038] Specifically: Assume the input feature is X∈R H×W×C . First, through three learnable weight matrices W Q， W K， W V map the input to the query, key, and value spaces: Q = XW Q , K = XW K , V = XW V .

[0039] Among them: Q, K, V∈R H×W×d (d is the attention dimension, usually , where d = 1 is taken here).

[0040] Calculate the attention weights: Calculate the attention scores through scaled dot product and apply Softmax normalization:

[0041] The denominator d is used to prevent the gradient instability caused by too large dot product values, and T is the transpose operation.

[0042] The output dimension is R H×W×d .

[0043] Residual connection and normalization: To retain the original information and stabilize the training, the self-attention output needs to be added to the input (residual connection), and then layer normalization is performed: Y = LayerNorm(X + Attention(Q, K, V)).

[0044] Residual connection: X + Attention(⋯).

[0045] Normalization: Normalize each channel of each sample, keeping the output and input in the same dimension.

[0046] Multi-layer perceptron: A multi-layer perceptron is a feedforward neural network composed of multiple linear layers and activation functions. It is used to perform non-linear transformations on features to extract more complex features. In the salinity intelligent inversion network model, the MLP is applied after the self-attention layer to further enhance the ability to represent brightness temperature features.

[0047] Layer normalization: Layer normalization is a normalization technique used to improve training stability and accelerate convergence. It normalizes the features of each sample so that their mean is 0 and standard deviation is 1. Different from batch normalization, layer normalization is performed within each sample rather than across batches.

[0048] Stochastic depth: Stochastic depth is a regularization technique used to enhance the generalization ability of the model. During training, some paths are randomly discarded through stochastic depth to reduce the model's dependence on certain specific paths, thereby improving the robustness of the model.

[0049] Multi-scale Convolutional Layer: To fully exploit the local feature information of the brightness temperature data at different scales, a multi-scale convolutional structure is introduced after the encoder module. Considering the continuity and local correlation of the brightness temperature data in the time series, a one-dimensional convolutional layer is selected as the basic building block, and different convolutional kernel sizes (3, 5, 7) are set to parallelly extract the change features of the brightness temperature data in the short-term, medium-term, and long-term ranges. All convolutional layers use a stride of 1 and a padding method to keep the output sequence length unchanged, and the unified output channel number is C = 64.

[0050] Specifically, the shape of the input high-dimensional brightness temperature feature data is [H, W / P, M], where W / P = 3 can correspond to the sizes of different convolutional kernels (3, 5, 7). To adapt to the input requirements of the one-dimensional convolution, the feature dimensions are first transposed to [H, M, W / P], and then the features at different scales are extracted by sliding multiple convolutional kernels in the time dimension. The convolution operation is uniformly set with a stride of 1 and boundary padding to keep the sequence length unchanged, and the output channel number is set to C = 64 to ensure the dimensional consistency of the output features. Finally, the local information of each scale is integrated by feature concatenation or weighted fusion to obtain a multi-scale brightness temperature feature representation with an output shape of [H, C, W / P].

[0051] Compared with the traditional convolutional layer and encoder, the multi-scale convolutional layer can provide richer local perception ability of the time series for subsequent salinity inversion, improve the sensitivity and discriminability of the model to the brightness temperature changes at different scales, and make the result of salinity inversion more accurate.

[0052] In the salinity intelligent inversion network, the role of the output layer is to map the high-dimensional feature representation of the brightness temperature data obtained from the encoder layer to the final prediction result or target value. In the specific implementation, it mainly consists of four parts: feature integration, MLP head (multi-layer perceptron), activation function, and linear transformation.

[0053] Feature Integration: The task of feature integration is to flatten the output of the encoder and map it to a vector of a fixed length through a linear transformation, converting the multi-dimensional features output by the encoding layer into a one-dimensional vector suitable for processing by a fully connected network. The output of the encoder with a shape of [H, C, W / P] is flattened into a tensor with a shape of [H, C×W / P] through a reshape operation for subsequent processing.

[0054] MLP Head (Multi-layer Perceptron): High-order features are gradually extracted through multi-layer non-linear transformations and finally mapped to the target output space. In the fully connected network part, 7 fully connected layers are defined and the features are fused through residual connections. The hidden layer dimensions of the 7 fully connected layers are all 128. The output of each layer is added to the sum of the outputs of all previous layers to retain the multi-scale information and alleviate the vanishing gradient.

[0055] Activation function: Introduce non-linearity to enhance the model's ability to fit complex patterns. Apply the ReLU activation function after each fully connected layer (except the last layer).

[0056] Linear transformation: Compress the features after the first 6 fully connected layers and residual connections into the low-dimensional salinity space required for the task.

[0057] Output layer These structures enable the model to effectively map the high-dimensional feature brightness temperature data obtained from the encoder to the salinity value results inverted by the final inversion model.

[0058] The method for obtaining the weight matrix is as follows: Obtain the mean, variance, maximum brightness temperature value for each month, and minimum brightness temperature value for each month of the low-dimensional brightness temperature feature data; Substitute the mean, variance, maximum brightness temperature value for each month, and minimum brightness temperature value for each month into the weight matrix formula to obtain the weight matrix; The weight matrix formula is: ; where f is the mapping function, is the mean, is the variance, max is the maximum brightness temperature value for each month, min is the minimum brightness temperature value for each month, is the weight matrix.

[0059] Obtaining the dynamic gating factor includes: Obtain the mean brightness temperature of the ocean surface for the current month; Obtain the average brightness temperature of the patch data; According to the mean brightness temperature, average brightness temperature, and the judgment formula, obtain the dynamic gating factor; The judgment formula is: ; where H is the number of grid points, W is the time length of the constructed brightness temperature dataset, C represents the number of channels of the brightness temperature dataset, P is the segmentation length, represents the average brightness temperature value of the original brightness temperature observations in the time dimension within the W / P-th time patch at grid point H, is the average brightness temperature, and are learning parameters, The value range of is 1.5 to 2.0, The value range of is 0.3 to 0.7.

[0060] The loss function is expressed as: ; where n represents the number of grid points in the target ocean area, represents the measured salinity value corresponding to the \(i\)-th grid point, represents the salinity value inverted by the inversion model corresponding to the \(i\)-th grid point of the model, and MAE is the value of the loss function.

[0061] S6. Input the actual brightness temperature data into the optimization model to obtain the salinity value.

[0062] The specific example is as follows: Figure 5 Shows the salinity data of August 2022 obtained by inverting with the brightness temperature data from June, July, and August 2015 to 2017 as the dataset and the buoy data as the label. Figure 6 Shows the salinity data of July 2022 obtained by inverting with the brightness temperature data from June, July, and August 2015 to 2017 as the dataset and the buoy data as the label. It can be seen from the figure that the salinity distribution in the South China Sea region obtained by inversion is similar to the actual salinity.

[0063] Figure 7 Reflects that the MSE values of the data of June, July, and August 2018 obtained by inverting with the data from June, July, and August 2015 to 2017 based on the brightness temperature-convolution collaborative coding network are 0.072, 0.1146, and 0.2113 respectively. Compared with using the data of this period as the training set to invert the data of 2019, 2020, 2021, and 2022, the MSE of the data of 2018 obtained by inversion is relatively the smallest. And with the increase of time, the value of MAE also shows an increasing trend. For example, using the data of July 2015 to 2017 as the training set, the MAE values of the data of July 2018 to 2022 obtained by inversion are 0.1146, 0.1341, 0.1880, 0.1955, and 0.2261 in turn. It can be seen that the accuracy of the sea surface salinity obtained by the inversion method used in this application is relatively high.

[0064] A sea surface salinity inversion system, comprising: An acquisition module, configured to acquire the training daily brightness temperature data and training monthly salinity data of the sea surface; A preprocessing module, configured to perform preprocessing on the training daily brightness temperature data and then perform regional screening to obtain regional brightness temperature data; Perform interpolation processing on the training monthly salinity data to obtain interpolated salinity data; An interpolation module, configured to align the regional brightness data and the interpolated salinity data in time to obtain a brightness temperature dataset and a salinity dataset; An optimization module, which is used to extract features from the brightness temperature dataset to obtain brightness temperature features, input the brightness temperature features into the salinity intelligent inversion network model to obtain an output salinity value, calculate the output loss value between the output salinity value and the salinity dataset through a loss function, and adjust the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; An output module, which is used to input actual brightness temperature data into the optimized model to obtain a salinity value.

[0065] An embodiment of the present application also discloses a terminal device, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, a salinity inversion method for the ocean surface is adopted.

[0066] Among them, the terminal device can be a computer device such as a desktop computer, a laptop computer or a cloud server. And the terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices and a bus, etc.

[0067] Among them, the processor can adopt a central processing unit (CPU). Of course, according to the actual usage situation, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc. The present application does not make any restrictions on this.

[0068] Among them, the memory can be an internal storage unit of the terminal device. For example, the hard disk or memory of the terminal device, or it can also be an external storage device of the terminal device. For example, a plug-in hard disk, a smart media card (SMC), a secure digital card (SD) or a flash card (FC) etc. equipped on the terminal device. And the memory can also be a combination of the internal storage unit and the external storage device of the terminal device. The memory is used to store the computer program and other programs and data required by the terminal device. The memory can also be used to temporarily store the data that has been output or will be output. The present application does not make any restrictions on this.

[0069] Among them, through this terminal device, a salinity inversion method for the ocean surface in the above embodiment is stored in the memory of the terminal device, and is loaded and executed on the processor of the terminal device, which is convenient for use.

[0070] An embodiment of the present application also discloses a computer-readable storage medium. And the computer-readable storage medium stores a computer program. Among them, when the computer program is executed by a processor, a salinity inversion method for the ocean surface in the above embodiment is adopted.

[0071] Among them, the computer program can be stored in a computer-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some middleware form, etc. The computer-readable medium includes any entity or device, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above components.

[0072] Among them, through this computer-readable storage medium, a method for salinity inversion on the ocean surface in the above embodiment is stored in the computer-readable storage medium, and is loaded and executed on a processor to facilitate the storage and application of the above method.

[0073] Those of ordinary skill in the art should understand that: The discussion of any of the above embodiments is only exemplary and is not intended to imply that the protection scope of this application is limited to these examples; under the idea of this application, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of one or more embodiments in this application as described above, and they are not provided in detail for the sake of brevity.

[0074] One or more embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application shall be included within the protection scope of this application.

Claims

1. A method for salinity inversion of the ocean surface, characterized in that Including: Obtain the training daily brightness temperature data and training monthly salinity data of the ocean surface; After preprocessing the training daily brightness temperature data, perform regional screening to obtain regional brightness temperature data; Perform interpolation processing on the training monthly salinity data to obtain interpolated salinity data; Align the regional brightness data and the interpolated salinity data in time to obtain a brightness temperature dataset and a salinity dataset; Extract features from the brightness temperature dataset to obtain brightness temperature features, input the brightness temperature features into the salinity intelligent inversion network model to obtain an output salinity value, calculate the output loss value between the output salinity value and the salinity dataset through a loss function, and adjust the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; Input the actual brightness temperature data into the optimized model to obtain a salinity value.

2. The salinity inversion method for the ocean surface according to claim 1, wherein, The step of performing interpolation processing on the training monthly salinity data to obtain interpolated salinity data includes: According to the training monthly salinity data, obtain the target coordinate value, the salinity corresponding to the target coordinate value, the four coordinate values closest to the target coordinate, and the salinities corresponding to the four coordinate values; Substitute the target coordinate value, the salinity corresponding to the target coordinate value, the four coordinate values closest to the target coordinate, and the salinities corresponding to the four coordinate values into the interpolation formula to obtain interpolated salinity data; The interpolation formula is: Where x and y are the horizontal and vertical coordinates of the target coordinate value, , and are the four coordinate values around the target coordinate, is the interpolated salinity data, and S is the salinity corresponding to the four coordinate values.

3. The salinity inversion method for the ocean surface according to claim 1, characterized in that, The step of performing regional screening on the training daily brightness temperature data after preprocessing to obtain regional brightness temperature data includes: Convert the format of the training daily brightness temperature data to obtain converted data; Correct the converted data to obtain corrected data; Set flag bits, where the flag bits include an ocean flag bit and other flag bits, the ocean flag bit is 1, and other flag bits are 0; Multiply the corrected data and the flag bits to obtain flag bit data; Obtain the screening area; Screen the flag bit data according to the screening area to obtain regional brightness temperature data.

4. The salinity inversion method for the ocean surface according to claim 1, characterized in that, The step of extracting features from the brightness temperature dataset to obtain brightness temperature features includes: Divide the brightness temperature dataset into time series to obtain a number of patch data, perform one-dimensional convolution operation on the patch data to obtain convolution data, and the brightness temperature dataset is represented as: H×W×C; Among them, H represents the number of grid points for constructing the brightness temperature dataset, W represents the time length of the constructed brightness temperature dataset, which is 30 days; C represents the number of channels of the brightness temperature dataset; Map the convolution data to a fixed embedding dimension D through a linear projection layer to obtain low-dimensional brightness temperature feature data; Input the low-dimensional brightness temperature feature data into a fully connected layer, a self-attention layer, a position encoding layer, and a normalization layer in sequence to obtain high-dimensional brightness temperature data, and use the high-dimensional brightness temperature data as the brightness temperature feature; The fully connected layer is represented as: ； Among them, X is the low-dimensional brightness temperature data output by the patch embedding layer and the position encoding layer, with a shape of [H, W / P, D], and Z is the high-dimensional brightness temperature data output after passing through the fully connected layer, with a shape of [H, W / P, M]. is a non-linear activation function, b is the learnable weight bias of the fully connected layer. is the learnable weight matrix of the fully connected layer, and the shape of the matrix is [D, M]. is the dynamic gating factor based on the ocean physics prior, and M is the dimension of the high-dimensional brightness temperature data.

5. The salinity inversion method for the ocean surface according to claim 4, characterized in that, The method for obtaining the weight matrix is: Obtain the mean, variance, the maximum brightness temperature value of each month, and the minimum brightness temperature value of each month of the low-dimensional brightness temperature feature data; Substitute the mean, variance, the maximum brightness temperature value of each month, and the minimum brightness temperature value of each month into the weight matrix formula to obtain the weight matrix; The weight matrix formula is: ； where f is a mapping function, is the mean value, is the variance, max is the maximum brightness temperature value of each month, and min is the minimum brightness temperature value of each month. is the weight matrix.

6. The method for salinity inversion of the ocean surface according to claim 4, characterized in that Obtaining the dynamic gating factor includes: Obtain the mean brightness temperature of the ocean surface in the current month; Obtain the average brightness temperature of the patch data; A dynamic gating factor is obtained according to the average brightness temperature, the mean brightness temperature, and the judgment formula; The judgment formula is: ； Among them, H is the number of grid points, W is the time length of the constructed brightness temperature dataset, C represents the number of channels of the brightness temperature dataset, and P is the segmentation length. denotes the average brightness temperature value of the original brightness temperature observations in the time dimension within the W / P-th time patch at grid point H. is the average brightness temperature. and are learning parameters. The value range of is 1.5 to 2.

0. The value range of is 0.3 to 0.

7.

7. The salinity inversion method for the ocean surface according to claim 1, wherein The loss function is expressed as: ； Among them, n represents the number of grid points in the target ocean area, represents the measured salinity value corresponding to the i-th grid point, represents the salinity value inverted by the inversion model corresponding to the i-th grid point of the model, and MAE is the value of the loss function.

8. A salinity inversion system for the ocean surface, characterized in that It includes: An acquisition module for acquiring the training-day brightness temperature data and the training-month salinity data of the ocean surface; A preprocessing module for preprocessing the training-day brightness temperature data and then performing regional screening to obtain regional brightness temperature data; Performing interpolation processing on the training-month salinity data to obtain interpolated salinity data; An interpolation module for time-aligning the regional brightness data and the interpolated salinity data to obtain a brightness temperature data set and a salinity data set; An optimization module for extracting features from the brightness temperature data set to obtain brightness temperature features, inputting the brightness temperature features into a salinity intelligent inversion network model to obtain an output salinity value, calculating an output loss value of the output salinity value and the salinity data set through a loss function, and adjusting the model parameters of the salinity intelligent inversion network model according to the output loss value to obtain an optimized model; An output module for inputting actual brightness temperature data into the optimized model to obtain a salinity value.

9. A terminal device, comprising a memory and a processor, characterized in that, The memory stores a computer program capable of running on a processor. When the processor loads and executes the computer program, the method described in any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium storing a computer program therein, characterized in that, When the computer program is loaded and executed by the processor, the method described in any one of claims 1 to 7 is adopted.

Images