A sea clutter shape parameter estimation method based on attention convolutional neural network

Abstract

This invention discloses a method for estimating sea clutter shape parameters based on an attention-based convolutional neural network. First, several sets of sea clutter distributions with the same scale parameters but different shape parameters are obtained. Then, based on the ratio of quantile features and the ratio of moment features for each set of sea clutter distributions, the joint "quantile-moment" features of that sea clutter distribution are determined. Using the scale parameters and the joint "quantile-moment" features of the sea clutter distribution as data input and the shape parameters as data output, a dataset is constructed. A convolutional neural network sea clutter shape parameter estimation model based on an attention mechanism is then built. The model is trained using the dataset, and the sea clutter shape parameters are estimated based on the trained model. This invention simultaneously considers the quantile and moment features of the sea clutter distribution, solving the problem of single features in traditional methods. Furthermore, by introducing an attention mechanism and a convolutional neural network, it can capture multi-dimensional and deep-level intrinsic correlations in the input features, compensating for the shortcomings of traditional methods in nonlinear tasks and providing support for marine information processing.

Description

Technical Field

[0001] This invention belongs to the field of sea clutter parameter estimation, specifically relating to a method for estimating sea clutter shape parameters based on attention convolutional neural networks. Background Technology

[0002] Modern radars need to filter out valid and invalid information before processing it. In maritime information processing technology, sea clutter is a typical example of invalid information and needs to be identified and suppressed during the data preprocessing stage. Therefore, to improve the information processing capabilities of radars, a more accurate understanding of the distribution characteristics of sea clutter is required.

[0003] Existing research has shown that the Generalized Pareto Distribution (GPD) can accurately fit the distribution of sea clutter. The GPD is determined by two parameters—a scale parameter and a shape parameter. The scale parameter describes the amplitude of the sea clutter distribution, while the shape parameter reflects the "long tail" effect of the sea clutter distribution. Therefore, accurately estimating the other parameter given either the scale or shape parameter is one of the important directions in studying the characteristics of sea clutter distribution.

[0004] Traditional methods for estimating sea clutter parameters include rule-based moment estimation and quantile estimation, statistical maximum likelihood estimation, and machine learning-based linear regression and decision tree models. These methods either utilize a limited number of features or employ models that lack nonlinear expressive power, thus exhibiting certain limitations in sea clutter parameter estimation. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method for estimating the shape parameters of sea clutter based on attention-based convolutional neural networks. It focuses on estimating the shape parameters of the generalized Pareto distribution of sea clutter under fixed scale parameters. Utilizing the attention mechanism and convolutional neural network technology in artificial intelligence, this invention proposes a multi-feature, nonlinear shape parameter estimation method for the generalized Pareto distribution of sea clutter under fixed scale parameters. This method can integrate multi-dimensional quantile features and moment features to capture the relationship between the shape parameters and these features, accurately estimating the values ​​of the shape parameters and providing support for sea clutter simulation and suppression research.

[0006] The specific technical solution for achieving the objective of this invention is as follows:

[0007] A method for estimating the shape parameters of sea clutter based on an attention-based convolutional neural network includes the following steps:

[0008] Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method.

[0009] Step 2: Determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0010] Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets;

[0011] Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism;

[0012] Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] (1) The solution of the present invention considers both the quantile characteristics and moment characteristics of the sea clutter distribution, which solves the problem of single characteristics in traditional methods. In addition, by introducing attention mechanism and convolutional neural network, it can capture the multi-dimensional and deep internal correlation in the input features, make up for the shortcomings of traditional methods in nonlinear tasks, and provide support for sea information processing.

[0015] (2) In view of the problems existing in the existing technical solutions, this invention utilizes the attention mechanism and convolutional neural network technology in artificial intelligence technology to propose a multi-feature, nonlinear shape parameter estimation method for the generalized Pareto distribution of sea clutter under fixed scale parameters. This method can integrate multi-dimensional quantile features and moment features, capture the relationship between shape parameters and these features, and accurately estimate the value of shape parameters.

[0016] The present invention will be further described below with reference to specific embodiments. Attached Figure Description

[0017] Figure 1 A schematic diagram of the process for estimating the shape parameters of sea clutter based on an attention convolutional neural network according to the present invention.

[0018] Figure 2 A schematic diagram of the attention-based convolutional neural network of the present invention. Detailed Implementation

[0019] A method for estimating the shape parameters of sea clutter based on an attention-based convolutional neural network includes the following steps:

[0020] Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method.

[0021] Step 2: Determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0022] The quantile feature ratio PF is:

[0023]

[0024] Among them, {P i1 ,P i2 ,P i3 ,…,P in ,P j1 ,P j2 ,P j3 ,…,P jm} represents multiple quantiles for each defined group of sea clutter distributions, where j1…j m ,i1…i n All are real numbers in the interval (0,1);

[0025] The moment characteristic ratio MF is:

[0026]

[0027] Among them, M n Let n denote the nth moment, n∈{0.5}∪N + ;

[0028] The combined "quantile-moment" characteristics of the sea clutter distribution are as follows:

[0029]

[0030] Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets;

[0031] The dataset can be divided into training, validation, and test sets in a 7:2:1 ratio.

[0032] Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism;

[0033] The inputs and outputs of the sea clutter shape parameter estimation model are as follows:

[0034] Input: [η, MPF]

[0035] Output: v

[0036] Where η is the scale parameter of the sea clutter distribution, and ν is the shape parameter of the sea clutter distribution.

[0037] The sea clutter shape parameter estimation model includes an attention mechanism and a convolutional neural network;

[0038] The attention mechanism utilizes the parameter matrix W to input the model. q W k and W v The linear transformation is applied to three distinct spaces Q, K, and V, and then the output of the attention mechanism is calculated using the following formula:

[0039]

[0040] Where, d k It is the scaling factor;

[0041] Next, the shape parameter η in the input is concatenated with the attention output Attn to obtain the final encoding result: Input = Concat(η, Attn);

[0042] The output of the attention mechanism, after passing through a convolutional neural network, uses the output of the fully connected layer as the probability distribution of the shape parameter estimates, and then uses the ArgMax function to calculate the final estimate v. pred ;

[0043] Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

[0044] The present invention also provides a method for estimating the shape parameters of sea clutter based on an attention convolutional neural network, characterized by comprising the following modules:

[0045] Dataset module: Used to preprocess sea clutter data that has been fitted with generalized Pareto, and to obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method to construct a dataset;

[0046] Feature extraction module: used to determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0047] Model building and training module: used to train the constructed attention-based convolutional neural network sea clutter shape parameter estimation model based on the dataset with extracted features;

[0048] Sea clutter shape parameter estimation module: Used to estimate sea clutter shape parameters using the trained model.

[0049] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0050] Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method.

[0051] Step 2: Determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0052] Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets;

[0053] Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism;

[0054] Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

[0055] The present invention also provides a computer-storable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the following steps:

[0056] Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method.

[0057] Step 2: Determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0058] Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets;

[0059] Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism;

[0060] Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

[0061] Example

[0062] To clearly describe the technical solution and effects achieved by the present invention, the technical solution of the present invention will be clearly explained below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can realize the invention without creative effort. The structure shown in the drawings is not the entirety of the actual structure but only a part of the actual structure. It should be noted that all other embodiments made by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the protection scope of the present invention.

[0063] The following examples are merely illustrative of the invention, and the scope of the invention is not limited to the examples provided. Therefore, any non-essential modifications made by those skilled in the art to the embodiments described above, applied to other embodiments, are still within the scope of protection of this invention. Furthermore, experimental methods not specified in the following examples should be performed according to conventional or manufacturer-recommended conditions. Unless otherwise specified, expressions in the text are used for distinguishing purposes only and have no other meaning.

[0064] Combination Figure 1 A method for estimating the shape parameters of sea clutter based on attention convolutional neural networks includes the following steps:

[0065] Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method.

[0066] Step 2: Determine the "quantile-moment" joint feature MPF of each sea clutter distribution based on the quantile feature ratio PF and moment feature ratio MF of each group of sea clutter distributions;

[0067] In this embodiment, the quantile feature ratio PF is:

[0068]

[0069] Among them, {P 0.2 ,P 0.25 ,P 0.3 ,P 0.35 ,P 0.5 ,P 0.55 ,P 0.6 ,P 0.65 ,P 0.7 ,P 0.75 ,P 0.8 ,P 0.85 ,P 0.9} represents the characteristics of multiple quantiles in each defined group of sea clutter distributions;

[0070] The moment characteristic ratio MF is:

[0071]

[0072] Among them, {M 0.5 {M1, M2, M3, M4} represent the moment characteristics of each defined group of sea clutter distributions, M... i The i-th moment of the sea clutter;

[0073] The combined "quantile-moment" characteristics of the sea clutter distribution are as follows:

[0074]

[0075] Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets;

[0076] The dataset can be divided into training, validation, and test sets in a 7:2:1 ratio.

[0077] Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism;

[0078] The inputs and outputs of the sea clutter shape parameter estimation model are as follows:

[0079] Input: [η, MPF]

[0080] Output:ν

[0081] Where η is the scale parameter of the sea clutter distribution, and ν is the shape parameter of the sea clutter distribution.

[0082] Combination Figure 2 The sea clutter shape parameter estimation model includes an attention mechanism and a convolutional neural network;

[0083] The attention mechanism utilizes the parameter matrix W to input the model. q W k and W v The linear transformation is applied to three distinct spaces Q, K, and V, and then the output of the attention mechanism is calculated using the following formula:

[0084]

[0085] Where, d k It is the scaling factor;

[0086] Next, the shape parameter η in the input is concatenated with the attention output Attn to obtain the final encoding result: Input = Concat(η, Attn);

[0087] The convolutional neural network in this embodiment is as follows: Convolutional layer 1 parameter settings: input channel 1, output channel 32, kernel size 3, stride 1;

[0088] Convolutional layer 2 parameter settings: input channels 32, output channels 64, kernel size 3, stride 1;

[0089] Fully connected layer 1 parameter settings: input dimension 256, output dimension 256;

[0090] Fully connected layer 2 parameter settings: input dimension 256, output dimension 200.

[0091] Finally, the output of fully connected layer 2 is used as the probability distribution of the shape parameter estimates [p] 0.1 ,p 0.2 ,p 0.3 ,...,p 20.0 The final estimated value v is calculated using the ArgMax function. pred .

[0092] Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

[0093] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for estimating the shape parameters of sea clutter based on an attention-based convolutional neural network, characterized in that, Includes the following steps: Step 1: Preprocess the sea clutter data that has been fitted with generalized Pareto, and obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method. Step 2: Based on the quantile characteristic ratio of each group of sea clutter distributions Sum of moment eigenvalues This allows for the determination of the joint "quantile-moment" characteristics of the sea clutter distribution. ; Step 3: Using the scale parameters and quantile-moment joint features of the sea clutter distribution as data inputs and the shape parameters as data outputs, construct training, validation, and test sets; Step 4: Construct a convolutional neural network-based sea clutter shape parameter estimation model based on an attention mechanism; Step 5: Based on the constructed model, train it using the dataset, and estimate the shape parameters of sea clutter based on the trained model.

2. The method for estimating sea clutter shape parameters based on attention convolutional neural networks according to claim 1, characterized in that, The quantile feature ratio in step 2 for: ； This represents multiple quantiles for each defined group of sea clutter distributions, where All are real numbers in the interval (0, 1).

3. The method for estimating sea clutter shape parameters based on attention convolutional neural networks according to claim 1, characterized in that, The moment characteristic ratio in step 2 for: ； in, express Step moment, .

4. The method for estimating sea clutter shape parameters based on an attention convolutional neural network according to claim 1 or 2, characterized in that, The "quantile-moment" joint characteristic of the sea clutter distribution in step 2 is as follows: 。 5. The method for estimating sea clutter shape parameters based on attention convolutional neural networks according to claim 1, characterized in that, The inputs and outputs of the sea clutter shape parameter estimation model in step 4 are as follows: ； in, The scale parameter for sea clutter distribution. The shape parameters of the sea clutter distribution.

6. The method for estimating sea clutter shape parameters based on an attention convolutional neural network according to claim 5, characterized in that, The sea clutter shape parameter estimation model includes an attention mechanism and a convolutional neural network; The attention mechanism utilizes the parameter matrix to input the model. , and Linear transformation to three different spaces , and Then, the output of the attention mechanism is calculated using the following formula: ； in, It is the scaling factor; The output of the attention mechanism is passed through a convolutional neural network. The output of the fully connected layer is used as the probability distribution of the shape parameter estimates, and the ArgMax function is used to calculate the final estimates. .

7. A system for estimating the shape parameters of sea clutter based on an attention-based convolutional neural network, characterized in that, Includes the following modules: Dataset module: Used to preprocess sea clutter data that has been fitted with generalized Pareto, and to obtain several sets of sea clutter distributions with the same scale parameter but different shape parameters based on the logistic method to construct a dataset; Feature extraction module: used to extract features based on the quantile feature ratios of each group of sea clutter distributions. Sum of moment eigenvalues This allows for the determination of the joint "quantile-moment" characteristics of the sea clutter distribution. ; Model building and training module: This module is used to train the attention-based convolutional neural network sea clutter shape parameter estimation model based on the dataset with extracted features. It uses the scale parameter of the sea clutter distribution and the joint features of "quantile-moment" as data input and the shape parameter as data output to build training, validation and test sets. Sea clutter shape parameter estimation module: Used to estimate sea clutter shape parameters using the trained model.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

9. A computer-storable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images