Sea clutter simulation method and device based on graph neural network

Abstract

The invention provides a sea clutter simulation method and device based on a graph neural network. The method comprises the steps of obtaining sea clutter data at a current moment; inputting the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain sea clutter prediction data; wherein the prediction data of the sea clutter is space-time distribution data of the sea clutter in a future time step length; the sea clutter prediction model comprises a graph learning module and a plurality of time-space convolution modules which are connected in sequence, and each time-space convolution module is composed of a time convolution layer and a graph convolution layer which are connected in sequence. The time convolution layer and the graph convolution layers sequentially extract time features and space features of sea clutters by capturing dependency relationships among nodes in historical sea clutter data, and the graph learning module is used for calculating a graph adjacency matrix according to node embedding information and inputting the graph adjacency matrix into each graph convolution layer. According to the scheme, accurate prediction simulation of the space-time distribution of the sea clutter in the future time step can be realized.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of signal and information processing, and particularly to a sea clutter simulation method and device based on a graph neural network. Background Art

[0003] In the related art, the model parameters are usually updated according to the current input data and the parameters in the current model. Sea clutter has spatio-temporal correlation along the time dimension and the distance dimension. However, this method cannot utilize the historical information of the sea clutter amplitude, so it can only understand the short-term change characteristics of the sea clutter and cannot study the long-term change law and spatio-temporal correlation of the sea clutter.

[0004] Therefore, there is an urgent need to provide a sea clutter simulation method and device based on a graph neural network. Summary of the Invention

[0005] In order to solve the problem that the traditional sea clutter simulation method cannot study the long-term change law and spatio-temporal correlation of sea clutter, and thus cannot accurately simulate the sea clutter within a period of time, the embodiments of the present invention provide a sea clutter simulation method and device based on a graph neural network.

[0006] In a first aspect, the embodiments of the present invention provide a sea clutter simulation method based on a graph neural network, including:

[0007] Obtain sea clutter data at the current moment;

[0008] Input the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain predicted data of the sea clutter; wherein, the predicted data of the sea clutter is the spatio-temporal distribution data of the sea clutter within a future time step.

[0009] The sea clutter prediction model includes a graph learning module and a plurality of spatio-temporal convolution modules connected in sequence. Each spatio-temporal convolution module is composed of a time convolution layer and a graph convolution layer connected in sequence. The time convolution layer and the graph convolution layer capture the dependence relationship between nodes in the historical sea clutter distribution map to sequentially extract the time feature and space feature of the sea clutter. The graph learning module is used to calculate the graph adjacency matrix according to the node embedding information and input it into each graph convolution layer.

[0010] In a second aspect, the embodiments of the present invention further provide a sea clutter simulation device based on a graph neural network, including:

[0011] An acquisition unit, configured to acquire sea clutter data at the current moment;

[0012] A prediction unit, configured to input the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain prediction data of sea clutter; wherein, the prediction data of sea clutter is spatio-temporal distribution data of sea clutter within a future time step.

[0013] The sea clutter prediction model includes a graph learning module and a plurality of sequentially connected spatio-temporal convolutional modules. Each spatio-temporal convolutional module is composed of a sequentially connected temporal convolutional layer and a graph convolutional layer. The temporal convolutional layer and the graph convolutional layer capture the dependencies between nodes in the historical sea clutter distribution map to sequentially extract the temporal features and spatial features of sea clutter. The graph learning module is configured to calculate a graph adjacency matrix based on node embedding information and input it into each graph convolutional layer.

[0014] In a third aspect, an embodiment of the present invention further provides a computing device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0015] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed on a computer, the computer is made to execute the method described in any embodiment of this specification.

[0016] On the other hand, an embodiment of the present application further provides a computer program product. The computer program product includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method described in any of the above embodiments.

[0017] An embodiment of the present invention provides a sea clutter simulation method based on a graph neural network. By setting a graph learning module and a plurality of sequentially connected spatio-temporal convolutional modules in a sea clutter prediction model, wherein each spatio-temporal convolutional module is composed of a sequentially connected temporal convolutional layer and a graph convolutional layer, this module can capture the dependencies between nodes in the historical sea clutter distribution map to respectively capture the temporal and spatial dependencies of sea clutter, and the graph learning module can adaptively extract a sparse graph adjacency matrix based on the input sea clutter data and input it into the graph convolutional layer, avoiding the over-smoothing problem that often occurs in the graph convolutional layer. Thus, by inputting the sea clutter data at the current moment into the above-constructed sea clutter prediction model, accurate prediction and simulation of the spatio-temporal distribution of sea clutter within a future time step are achieved. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of a sea clutter simulation method based on a graph neural network provided by an embodiment of the present invention;

[0020] Figure 2 It is a schematic structural diagram of a sea clutter prediction model provided by an embodiment of the present invention;

[0021] Figure 3 It is a prediction result graph and a measured graph of sea clutter of a sea clutter simulation method based on a graph neural network provided by an embodiment of the present invention; among them, the left side is the measured sea clutter data, and the right side is the predicted sea clutter data;

[0022] Figure 4 It is a hardware architecture diagram of a computing device provided by an embodiment of the present invention;

[0023] Figure 5 It is a structural diagram of a sea clutter simulation device based on a graph neural network provided by an embodiment of the present invention. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0025] The following describes the specific implementation manners of the above concepts.

[0026] Please refer to Figure 1 , an embodiment of the present invention provides a sea clutter simulation method based on a graph neural network, and the method includes:

[0027] Step 100, obtaining sea clutter data at the current moment;

[0028] Step 102, inputting the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain predicted data of sea clutter; wherein, the predicted data of sea clutter is the spatio-temporal distribution data of sea clutter within a future time step.

[0029] The sea clutter prediction model includes a graph learning module and a number of sequentially connected spatio-temporal convolution modules. Each spatio-temporal convolution module is composed of a sequentially connected temporal convolution layer and a graph convolution layer. The temporal convolution layer and the graph convolution layer capture the dependencies between nodes in the historical sea clutter data to sequentially extract the temporal features and spatial features of the sea clutter. The graph learning module is used to calculate the graph adjacency matrix based on the node embedding information and input it into each graph convolution layer.

[0030] In the embodiment of the present invention, by setting a graph learning module and a number of sequentially connected spatio-temporal convolution modules in the sea clutter prediction model, where each spatio-temporal convolution module is composed of a sequentially connected temporal convolution layer and a graph convolution layer, this module can capture the dependencies between nodes in the historical sea clutter data to respectively capture the temporal and spatial dependencies of the sea clutter, and the graph learning module can adaptively extract the sparse graph adjacency matrix based on the input sea clutter data and input it into the graph convolution layer, avoiding the over-smoothing problem that often occurs in the graph convolution layer. Thus, by inputting the sea clutter data at the current moment into the above-built sea clutter prediction model, an accurate prediction simulation of the spatio-temporal distribution of sea clutter in the future time step is realized.

[0031] Regarding step 100:

[0032] In the embodiment of the present invention, the IPIX technology can be used to obtain the ocean surface reflection signal data at the current moment to obtain the sea clutter data at the current moment. This technology can capture the influence of small fluctuations on the sea surface on the radar echo, thereby generating a high-resolution sea clutter image, and inputting the high-resolution sea clutter image into the sea clutter prediction model, using the relationship between nodes and edges in the sea clutter image to capture complex data features.

[0033] Regarding step 102:

[0034] In some embodiments, the sea clutter prediction model further includes an input module, a 1×1 convolution module, and an output module; wherein, the input module is connected to the 1×1 convolution module, the 1×1 convolution module is used to reduce the dimension and fuse features of the sea clutter data obtained by the input module, the 1×1 convolution module is connected to the spatio-temporal convolution module, and the spatio-temporal convolution module is jump-connected to the output module to output the predicted data of the sea clutter.

[0035] The sea clutter data has a certain correlation along the time dimension and the distance dimension. In order to accurately extract the spatio-temporal correlation of the sea clutter, the embodiment of the present invention constructs a sea clutter prediction model, which consists of Figure 2As can be seen, the sea clutter prediction model in the embodiments of the present invention includes an input module, a 1×1 convolution module, an output module, a graph learning layer module, and a number of connected spatio-temporal convolution modules. The spatio-temporal convolution module is formed by interleaving a graph convolution layer and a temporal convolution layer. The input module is used to obtain sea clutter data X∈R Tm×N×D , where T is the time step, N is the number of nodes, D is the feature dimension of each node. After receiving the sea clutter data obtained by the input module, the 1×1 convolution module reduces the dimension of this data, thereby effectively reducing the number of channels of the input sea clutter data feature map, reducing the complexity and computational amount of the model. Then, the output of the 1×1 convolution module enters the temporal convolution layer to process the temporal dependence in the sequence data to extract temporal features, and the extracted features are input into the graph convolution layer. The graph learning module adaptively calculates the graph adjacency matrix according to the node embedding information of the sea clutter data and inputs it into the graph convolution layer. The graph convolution layer receives the output data from the temporal convolution layer and the graph learning module and performs convolution processing on it to capture the spatial dependence relationship between nodes. Moreover, the residual connection directly adds the input of each temporal convolution layer to the output of the graph convolution layer after this layer, and then passes it to the next temporal convolution layer as its input, which helps to alleviate the problem of gradient disappearance in the deep network and allows the gradient to propagate more directly from the output layer back to the previous layers. In addition, each temporal convolution layer jumps before the graph learning layer in the next spatio-temporal convolution module and directly passes it to the output module, which helps to retain the direct path from the input to the output and improve the performance of the model. The above processing process fully considers the temporal and spatial features in the historical sea clutter data, thereby realizing the accurate simulation prediction of the sea clutter data in the future time period.

[0036] The skip connection is essentially a standard convolution of 1×N i , where N i represents the sequence length of the input of the i-th skip connection layer, and it normalizes the information that jumps to the output module to have the same sequence length of 1. The output module is composed of two standard convolution layers of 1×1, which convert the input channel dimension into the required output dimension. When only a single-step prediction is required, the output dimension is 1. When a continuous step length of length L needs to be predicted, the output dimension is L.

[0037] It should be noted that in the embodiments of the present invention, the number of spatio-temporal convolution modules is not specifically limited. Considering that as the number of modules increases, the model training time will become longer, so the appropriate number of spatio-temporal convolution modules can actually be selected according to the training requirements.

[0038] In some embodiments, the graph learning module is constructed in the following manner:

[0039] According to the node embedding matrices and model bias terms with two different features in the sea clutter data, two intermediate matrices M1 and M2 are calculated using the tanh activation function.

[0040] M1 = tanh(αE1Ο1)

[0041] M2 = tanh(αE2Ο2)

[0042] Where α is a scaling factor, E1 and E2 are respectively node embedding matrices with two different randomly initialized features, Ο1 and Ο2 are two different bias terms in the model, and tanh is the hyperbolic tangent activation function.

[0043] Calculate the difference of the outer products of the above two intermediate matrices M1 and M2, and process them successively using the tanh activation function and the RELU activation function to obtain the attention weight matrix A, and use this attention weight matrix as the graph adjacency matrix reflecting the association strength between nodes.

[0044]

[0045] For each node in the sea clutter data, select the maximum index in the attention weight matrix corresponding to each node, and set the value corresponding to this maximum index to zero to update the attention weight matrix.

[0046] for = 1, 2, …, N

[0047] idx = argtopk(A[i, :])

[0048] A[i, -idx] = 0,

[0049] Where i is a loop variable used to traverse each node of the sea clutter data, N is the total number of nodes, and idx is the index obtained through the argtopk function, indicating that for each node i, the indexes of the k largest values in the i-th row of the A matrix are selected and set to 0.

[0050] In the embodiments of the present invention, in order to extract the hidden features between nodes in the sea clutter data, through the attention weight matrix A, and use it as the graph adjacency matrix reflecting the association strength between nodes, and update the attention weight matrix according to the maximum index in the attention weight matrix corresponding to each node, clarify the most significant relevant nodes for each node, thereby highlighting the most important connection relationships, enabling the model to focus on the most critical features, improving the calculation efficiency, and enhancing the robustness and expression ability of the model.

[0051] In some embodiments, the temporal convolutional layer includes a set of extended one-dimensional convolutional filters for extracting high-level temporal features; wherein each extended one-dimensional convolutional filter is respectively connected to a tanh activation function and a sigmod activation function, and the sigmod activation function is used to control the amount of information passed by the extended one-dimensional convolutional filter to the next module.

[0052] In the embodiment of the present invention, the temporal convolutional layer uses a set of standard extended one-dimensional convolutional filters to extract high-level temporal features, which are specifically composed of two extended one-dimensional convolutional filters. One filter is connected to a tanh activation function, and the other filter is connected to a sigmod activation function. This activation function acts as a gate to control the amount of information passed by the filter to the next module. The two activation functions can introduce different non-linear characteristics, enabling the model to learn more complex function mappings and helping to improve the model's expressive ability.

[0053] In some embodiments, the temporal convolutional layer sequentially convolves the feature information output by the 1×1 convolutional module using convolutional kernels of sizes 1×2, 1×3, 1×6, and 1×7, and connects the outputs of different convolutional kernels.

[0054] In the embodiment of the present invention, the size of the convolutional kernel in the temporal convolutional layer has a great influence on the performance of the convolutional network. Choosing the correct size of the convolutional kernel is crucial. If the convolutional kernel is too large, it cannot subtly represent short-term signal patterns; if it is too small, it cannot fully discover long-term signal patterns. Considering that time signals tend to have several inherent periods, such as 7, 12, 24, 28, and 60, and the traditional sizes of 1×1, 1×3, and 1×5 convolutional kernels are not suitable for the nature of time signals and cannot well contain these periods. Therefore, in the embodiment of the present invention, a temporal initial layer composed of four convolutional kernels of sizes 1×2, 1×3, 1×6, and 1×7 is proposed. In this way, it can achieve full coverage of time signals and capture long-time sequence information. At the same time, using convolutional kernels of different sizes can capture features of different scales of the input data and connect the outputs of the convolutional kernels. Through the concatenation operation, the features extracted by different convolutional kernels can be merged, which can retain more information and help the model learn more complex feature representations, thus contributing to a more comprehensive analysis and prediction of the spatio-temporal variation law of sea clutter. Capturing time patterns through improved one-dimensional convolution can discover time patterns with multiple frequencies and can also process very long sequences.

[0055] Specifically, for a given sea clutter sequence z ∈ R T , f 1×2 ∈ R 2 , f 1×3 ∈ R 3 , f1×6 ∈R 6 , f 1×7 ∈R 7 , the mathematical expression of the convolution process of this temporal convolution layer is as follows:

[0056] z = concat(z * f 1×2 , z * f 1×3 , z * f 1×6 , z * f 1×7 )

[0057]

[0058] In the formula, R T is the time series data of dimension T, z is the element in the time series, and f 1×2 is a 1x2 convolution kernel, and d is the dilation factor.

[0059] In some embodiments, the processing process of the graph convolution module is as follows:

[0060]

[0061] In the formula, H k is the hidden state vector of a node in the k-th layer, λ is a hyperparameter that controls the ratio of retaining the original state of the root node, K is the propagation depth, H i is the input hidden state output by the previous layer, H o is the output hidden state of the current layer, H 0 = H i , is the normalized adjacency matrix, A is the original adjacency matrix, I is the identity matrix, D is the node degree matrix on the diagonal, and W k is the weight matrix of the k-th layer, and H (k-1) represents the hidden state vector of a node in the (k - 1)-th layer.

[0062] In some embodiments, the sea clutter prediction model is trained with historical sea clutter data as the input training sample set and the spatio-temporal distribution data of sea clutter within the future time step as the output training sample set.

[0063] In the embodiments of the present invention, in order to verify the accuracy of the sea clutter prediction model, the IPIX sea clutter data is used to train the sea clutter prediction model. Among them, the data set is divided into a training set, a validation set, and a test set, and the proportions of the three are 0.7, 0.2, and 0.1 respectively. During the entire training phase, the network structure is adjusted to gradually reduce the validation error. The model structure and weights that produce the minimum validation error will be retained.

[0064] The sea clutter prediction model is developed and implemented using the PyTorch framework. The sea clutter prediction model is trained through a neural network, operating at 3.7 GHz on the Windows 11 platform, and using an NVIDIA GTX 3060Ti graphics card to enhance the training process. Additionally, the input layer of the sea clutter prediction model requires a fixed size, so the input layer is adjusted to a compatible size for the network. The input of the sea clutter prediction model is set to 430×12, and the number of iterations and learning rate are set to 50 and 0.0001 respectively. The Dropout technique is utilized to facilitate the training of the deep neural network. By ignoring half of the feature detectors in each training batch, overfitting is effectively reduced. The Dropout rate is set to 0.5, and the batch size is set to 64.

[0065] The sea clutter prediction model obtained from the above training is used to predict the sea clutter. As can be seen from Figure 3 the simulation sea clutter data predicted in the embodiments of the present invention has the same spatio-temporal distribution as the actual sea clutter under the same measurement conditions, with high prediction accuracy. Moreover, the sea clutter prediction model has good robustness.

[0066] As Figure 4 、 Figure 5 shown, the embodiments of the present invention provide a sea clutter simulation device based on a graph neural network. The device embodiments can be implemented through software, or through hardware or a combination of software and hardware. From a hardware perspective, as Figure 4 shown, it is a hardware architecture diagram of a computing device where the sea clutter simulation device based on a graph neural network provided by the embodiments of the present invention is located. In addition to Figure 4 the processors, memory, network interfaces, and non-volatile memories shown, the computing devices where the devices in the embodiments are located usually may also include other hardware, such as forwarding chips responsible for processing packets, etc. Taking software implementation as an example, as Figure 5 shown, as a logically defined device, it is formed by the CPU of its corresponding computing device reading the computer program in the non-volatile memory into the memory for operation.

[0067] A sea clutter simulation device based on a graph neural network provided in this embodiment, the device includes:

[0068] An acquisition unit 501, configured to acquire sea clutter data at the current moment;

[0069] A prediction unit 502, configured to input the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain prediction data of the sea clutter; wherein, the prediction data of the sea clutter is spatio-temporal distribution data of the sea clutter within a future time step.

[0070] The sea clutter prediction model includes a graph learning module and a number of sequentially connected spatio-temporal convolution modules. Each spatio-temporal convolution module is composed of a sequentially connected temporal convolution layer and a graph convolution layer. The temporal convolution layer and the graph convolution layer capture the dependencies between nodes in the historical sea clutter data to sequentially extract the temporal features and spatial features of the sea clutter. The graph learning module is used to calculate the graph adjacency matrix based on the node embedding information and input it into each graph convolution layer.

[0071] In an embodiment of the present invention, the acquisition unit 501 can be used to execute step 100 in the above method embodiment, and the prediction unit 502 can be used to execute step 102 in the above method embodiment.

[0072] In an embodiment of the present invention, in the prediction unit 502, the sea clutter prediction model is trained with historical sea clutter data as the input training sample set and the spatio-temporal distribution data of sea clutter within the future time step as the output training sample set.

[0073] In an embodiment of the present invention, in the prediction unit 502, the sea clutter prediction model further includes an input module, a 1×1 convolution module, and an output module. Among them, the input module is connected to the 1×1 convolution module. The 1×1 convolution module is used to reduce the dimension of the sea clutter data obtained by the input module. The 1×1 convolution module is connected to the spatio-temporal convolution module, and the spatio-temporal convolution module is jump-connected to the output module to output the predicted data of the sea clutter.

[0074] In an embodiment of the present invention, in the prediction unit 502, the graph learning module is constructed in the following manner:

[0075] According to the node embedding matrices of two different features and the model bias term in the sea clutter data, two intermediate matrices are calculated using the tanh activation function.

[0076] Calculate the difference between the outer products of the above two intermediate matrices, and sequentially process them using the tanh activation function and the RELU activation function to obtain the attention weight matrix, and use this attention weight matrix as the graph adjacency matrix reflecting the association strength between nodes.

[0077] For each node in the sea clutter data, select the maximum index in the attention weight matrix corresponding to each node, and set the value corresponding to this maximum index to zero to update the attention weight matrix.

[0078] In an embodiment of the present invention, in the prediction unit 502, the temporal convolutional layer includes a set of extended one-dimensional convolutional filters for extracting high-level temporal features; wherein each extended one-dimensional convolutional filter is respectively connected to a tanh activation function and a sigmod activation function, and the sigmod activation function is used to control the amount of information passed by the extended one-dimensional convolutional filter to the next module.

[0079] In an embodiment of the present invention, in the prediction unit 502, the temporal convolutional layer sequentially performs convolution on the feature information output by the 1×1 convolution module using convolutional kernels of sizes 1×2, 1×3, 1×6, and 1×7, and connects the outputs of different convolutional kernels.

[0080] It can be understood that the structure schematically shown in the embodiments of the present invention does not constitute a specific limitation on a sea clutter simulation device based on a graph neural network. In other embodiments of the present invention, a sea clutter simulation device based on a graph neural network may include more or fewer components than those shown, or combine certain components, or split certain components, or have different component arrangements. The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0081] Regarding the information interaction, execution process, etc. between the various modules within the above-mentioned device, since they are based on the same concept as the method embodiments of the present invention, the specific content can be referred to the description in the method embodiments of the present invention and will not be elaborated here.

[0082] The embodiments of the present invention also provide a computing device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements a sea clutter simulation method based on a graph neural network in any embodiment of the present invention.

[0083] The embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor is caused to execute a sea clutter simulation method based on a graph neural network in any embodiment of the present invention.

[0084] Specifically, a system or device equipped with a storage medium can be provided, on which software program code for implementing the functions in any of the above embodiments is stored, and the computer (or CPU or MPU) of the system or device is caused to read and execute the program code stored in the storage medium.

[0085] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments, so the program code and the storage medium storing the program code constitute a part of the present invention.

[0086] Examples of storage media for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the program code can be downloaded from a server computer via a communication network.

[0087] In addition, it should be clear that not only can the functions of any of the above embodiments be realized by executing the program code read by a computer, but also by causing an operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.

[0088] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in an expansion board inserted into the computer or into the memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion module is caused to execute part or all of the actual operations, thereby realizing the functions of any of the above embodiments.

[0089] Embodiments of the present application also provide a computer-readable storage medium, on which at least one instruction, at least one program, a code set, or an instruction set is stored. The at least one instruction, at least one program, the code set, or the instruction set is loaded and executed by a processor to implement a method for simulating sea clutter based on a graph neural network provided by any of the above method embodiments.

[0090] Embodiments of the present application also provide a computer program product, which includes a computer program. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method for measuring the atmospheric air permeability based on an infrared device described in any of the above embodiments.

[0091] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0092] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including those of the above method embodiments. The aforementioned storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disks, or optical discs.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sea clutter simulation method based on graph neural network, characterized in that, including: Obtain the sea clutter data at the current moment; Input the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain the predicted data of the sea clutter; wherein, the predicted data of the sea clutter is the spatio-temporal distribution data of the sea clutter within a future time step; The sea clutter prediction model includes a graph learning module and a number of sequentially connected spatio-temporal convolutional modules. Each spatio-temporal convolutional module is composed of a sequentially connected temporal convolutional layer and a graph convolutional layer. The temporal convolutional layer and the graph convolutional layer capture the dependencies between nodes in the historical sea clutter data to sequentially extract the temporal features and spatial features of the sea clutter. The graph learning module is used to calculate the graph adjacency matrix based on the node embedding information and input it into each graph convolutional layer.

2. The method according to claim 1, wherein The sea clutter prediction model is trained with the historical sea clutter data as the input training sample set and the spatio-temporal distribution data of the sea clutter within a future time step as the output training sample set.

3. The method according to claim 1, wherein The sea clutter prediction model further includes an input module, a 1×1 convolutional module, and an output module; wherein, the input module is connected to the 1×1 convolutional module. The 1×1 convolutional module is used to reduce the dimension of the sea clutter data obtained by the input module. The 1×1 convolutional module is connected to the spatio-temporal convolutional module. The spatio-temporal convolutional module is skip-connected to the output module to output the predicted data of the sea clutter.

4. The method according to claim 1, wherein The graph learning module is constructed in the following manner: According to the node embedding matrices of two different features and the model bias term in the sea clutter data, use the tanh activation function to calculate two intermediate matrices; Calculate the difference between the outer products of the above two intermediate matrices, and sequentially process them using the tanh activation function and the RELU activation function to obtain the attention weight matrix, and use this attention weight matrix as the graph adjacency matrix reflecting the association strength between nodes; For each node in the sea clutter data, select the maximum index in the attention weight matrix corresponding to each node, and set the value corresponding to the maximum index to zero to update the attention weight matrix.

5. The method according to claim 1, wherein The temporal convolutional layer includes a set of extended one-dimensional convolutional filters for extracting high-level temporal features; wherein, each extended one-dimensional convolutional filter is respectively connected to the tanh activation function and the sigmod activation function. The sigmod activation function is used to control the amount of information passed by the extended one-dimensional convolutional filter to the next module.

6. The method according to claim 5, characterized in that The temporal convolutional layer sequentially convolves the feature information output by the 1×1 convolutional module with convolutional kernels of sizes 1×2, 1×3, 1×6, and 1×7, and connects the outputs of different convolutional kernels.

7. A sea clutter simulation device based on a graph neural network, characterized in that, including: An acquisition unit for acquiring the sea clutter data at the current moment; A prediction unit for inputting the sea clutter data at the current moment into a pre-constructed sea clutter prediction model to obtain the predicted data of the sea clutter; wherein, the predicted data of the sea clutter is the spatio-temporal distribution data of the sea clutter within a future time step; The sea clutter prediction model includes a graph learning module and a number of sequentially connected spatio-temporal convolutional modules. Each spatio-temporal convolutional module is composed of a sequentially connected temporal convolutional layer and a graph convolutional layer. The temporal convolutional layer and the graph convolutional layer capture the dependencies between nodes in the historical sea clutter data to sequentially extract the temporal features and spatial features of the sea clutter. The graph learning module is used to calculate the graph adjacency matrix based on the node embedding information and input it into each graph convolutional layer.

8. A computing device, comprising a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the method according to any one of claims 1-6 is implemented.

9. A computer-readable storage medium, having a computer program stored thereon. When the computer program is executed on a computer, the computer is made to execute the method according to any one of claims 1-6.

10. A computer program product, characterized in that, Including a computer program, which when executed by a processor implements the method according to any one of claims 1-6.

Images