A radar clutter suppression method and system based on diffusion model and low-rank matrix decomposition

Abstract

The present invention relates to the field of generative artificial intelligence and radar image processing technology, and in particular to a radar clutter suppression method and system based on a diffusion model and low-rank matrix decomposition. The method comprises acquiring radar signal data; preprocessing and numerically filling the radar signal data; simulating the diffusion process of the radar signal using a diffusion model, and extracting features of the radar signal to obtain a clutter image of the radar signal; reducing the dimension of the clutter image to a latent space by an autoencoder to obtain a noise image, and sequentially performing forward diffusion and reverse diffusion on the noise image to generate a clutter-free radar image; the present invention can effectively separate target signals and clutter components in radar images by low-rank matrix decomposition. Low-rank components mainly represent target signals, while sparse components capture clutter and noise. Through this separation, target information can be more accurately identified and processed.

Description

Technical Field

[0001] The present invention relates to the technical field of generative artificial intelligence and radar image processing, and in particular to a radar clutter suppression method and system based on a diffusion model and low-rank matrix decomposition. Background Art

[0002] Radar clutter suppression methods combining diffusion models and low-rank matrix decomposition techniques are increasingly used in modern radar signal processing, aiming to improve the target detection capability of radar systems.

[0003] Sea clutter is radar echo signals caused by waves, swells and other irregular movements on the sea surface, which can interfere with the radar system's detection of targets. Suppressing sea clutter helps improve the radar system's detection accuracy of targets such as ships, boats, buoys, etc., thereby avoiding false positives and false negatives. Sea clutter signals are often stronger and more widely distributed than target signals, which can mask or confuse target signals. By suppressing sea clutter, the radar system's ability to identify target signals can be enhanced, making the target more obvious and easier to identify in the radar image.

[0004] The SD (Stochastic Diffusion) diffusion model is a diffusion model based on random processes and is often used to simulate the diffusion behavior of signals in random and uncertain environments. In the field of radar clutter suppression, the SD diffusion model can be used to simulate and understand the complex spatial distribution characteristics of clutter, thereby improving the effect of clutter suppression.

[0005] The SD diffusion model can simulate the random diffusion behavior of clutter signals in space, which helps to understand the spatial distribution characteristics and change laws of clutter. By extracting the characteristics of clutter signals through the SD diffusion model, it is possible to better distinguish them from target signals, thereby improving the accuracy of clutter suppression. In radar clutter suppression, the SD diffusion model can be used in combination with low-rank matrix decomposition technology. First, the clutter characteristics are simulated and extracted through the SD diffusion model, and then the clutter is separated from the target signal using low-rank matrix decomposition technology. The SD diffusion model simulates the diffusion behavior of the signal by introducing a random process, which can effectively describe and extract the clutter characteristics. In radar clutter suppression, combined with low-rank matrix decomposition technology, the effect of clutter suppression can be improved and the target detection capability of the radar system can be enhanced. The SD diffusion model describes the diffusion process of the signal by introducing random noise, taking into account the uncertainty factors in the environment. This randomness can be represented by a stochastic differential equation (SDE). Classical diffusion models usually use partial differential equations to describe the diffusion process, such as the heat conduction equation. The SD diffusion model simulates random diffusion behavior by introducing random terms in these equations. Summary of the invention

[0006] In order to solve the problem that the sea clutter mentioned above has highly complex and random characteristics, it is difficult for traditional methods to accurately model and suppress it. The sea clutter signal has different characteristics at different spatial locations and time points, which is difficult to describe with a simple statistical model. The ocean environment is constantly changing, causing the clutter characteristics to change over time. The present invention provides a radar clutter suppression method and system based on a diffusion model and low-rank matrix decomposition. To solve the problem that the ocean environment is constantly changing, causing the clutter characteristics to change over time.

[0007] In a first aspect, the present invention provides a radar clutter suppression method using a diffusion model and low-rank matrix decomposition, which adopts the following technical solutions:

[0008] A radar clutter suppression method based on a diffusion model and low-rank matrix decomposition, comprising:

[0009] Acquire radar signal data;

[0010] Preprocess and fill in the radar signal data;

[0011] The diffusion model is used to simulate the diffusion process of radar signals, and the features of radar signals are extracted to obtain the clutter image of radar signals.

[0012] The clutter image is reduced to the latent space through the autoencoder to obtain the noisy image;

[0013] The noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image;

[0014] The clutter suppression of radar signal data is evaluated by loss function, SCNR and SINR.

[0015] Furthermore, the preprocessing and value filling of the radar signal data includes filtering the radar data to remove noise and unnecessary frequency components in the radar signal, wherein a mean filter is used to smooth the signal, which is expressed as:

[0016] y n is the signal value after filtering, x n+k is the value of the original signal in the window, M is the half width of the window, which determines the smoothness of the filter. n , output signal y n , the window size is 2M+1.

[0017] Furthermore, the preprocessing and value filling of the radar signal data further includes mirror filling of the radar signal beyond the window range at the signal boundary, which is expressed as:

[0018]

[0019] The input signal is x n , whose length is N,x -n is the value of the signal at position -n, applicable to the case where n<0, x 2N-n-2 It is the value of the signal at position 2N-n-2, applicable to the case where n≥N.

[0020] Furthermore, the use of the diffusion model to simulate the diffusion process of the radar signal includes using the diffusion model to simulate the random diffusion behavior of the clutter signal in the radar signal in space, wherein the diffusion model in discrete form is used and is expressed as:

[0021]

[0022] Where Δt is the time step, η(t,x) is a standard normally distributed random variable representing the random perturbation in the discrete time step, and the noise intensity , represents the intensity of random disturbance; D is the diffusion coefficient, C(t,x) represents the intensity of the clutter signal at time t and position x, and ▽2 is the Laplace operator, which represents the diffusion effect in space.

[0023] Furthermore, the feature extraction of the radar signal includes distinguishing the clutter signal and the target signal in the radar signal based on the diffusion coefficient, the noise lightness and the spatial distribution characteristics, extracting the diffusion feature by calculating the Laplace operator of the clutter signal, extracting the random feature by estimating the noise intensity and the contribution of the Wiener process, and finally obtaining the autocorrelation of the radar signal over time by simulating the distribution of the clutter signal.

[0024] Furthermore, the noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image, including converting the noise image x0 into a latent space feature representation, denoted as z0, and the forward diffusion process of the diffusion model is:

[0025]

[0026] Where zt is the feature representation at step t. βt is the noise weight at step t. εt is the noise sampled from a standard normal distribution, starting from t=0t=0t=0 and gradually increasing the noise until pure noise is generated.

[0027] Furthermore, the process of sequentially performing forward diffusion and reverse diffusion on the noise image to generate a clutter-free radar image also includes reverse diffusion of pure noise, which is expressed as:

[0028]

[0029] zt is the feature representation at step t, βt is the noise weight at step t, εt is the noise sampled from a standard normal distribution, and a noise prediction model is used Estimate the noise εt to reduce the impact of noise, from pure noise z T At the beginning, through multiple iterations, the noise is gradually reduced to generate the target image.

[0030] In the second aspect, a radar clutter suppression system based on a diffusion model and low-rank matrix decomposition includes:

[0031] A data acquisition module is configured to acquire radar signal data;

[0032] The preprocessing module is configured to preprocess and fill in the radar signal data;

[0033] The feature extraction module is configured to simulate the diffusion process of the radar signal using a diffusion model and extract features of the radar signal to obtain a clutter image of the radar signal;

[0034] The noise module is configured to reduce the dimension of the noise image to the latent space through the autoencoder to obtain the noise image;

[0035] The diffusion module is configured to perform forward diffusion and reverse diffusion on the noise image in sequence to generate a clutter-free radar image;

[0036] The evaluation module is configured to evaluate the clutter suppression of the radar signal data through the loss function, SCNR and SINR.

[0037] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, the radar clutter suppression method of a diffusion model and low-rank matrix decomposition.

[0038] In a fourth aspect, the present invention provides a terminal device, including a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the radar clutter suppression method of a diffusion model and low-rank matrix decomposition.

[0039] In summary, the present invention has the following beneficial technical effects:

[0040] (1) The present invention proposes a radar clutter suppression method that combines a diffusion model with low-rank matrix decomposition technology. Low-rank matrix decomposition can effectively separate target signals and clutter components in radar images. The low-rank components mainly represent the target signals, while the sparse components capture clutter and noise. Through this separation, target information can be more accurately identified and processed.

[0041] (2) The radar clutter suppression method proposed in this invention, which combines the diffusion model and low-rank matrix decomposition technology, can effectively improve the denoising effect. The diffusion model gradually adds and removes noise through multi-step iterations, which can effectively simulate and remove clutter and noise in the image. The noise prediction model in the inverse diffusion process can accurately restore the image and improve the denoising effect.

[0042] (3) The radar clutter suppression method proposed in this invention, which combines the diffusion model and low-rank matrix decomposition technology, can enhance the target detection and recognition capabilities. Through effective clutter suppression, the target signal in the radar image is more prominent, reducing the interference of clutter on target detection and recognition, thereby improving the image quality and the accuracy of target recognition.

[0043] (4) The radar clutter suppression method proposed in this invention, which combines the diffusion model and low-rank matrix decomposition technology, can adaptively model noise. The diffusion model can adaptively learn and simulate the characteristics of different types of noise, thereby showing good denoising performance under different ocean and meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the process framework of Example 1 of the present invention.

[0045] Figure 2 It is a time-frequency spectrum comparison diagram of sea clutter suppression of different methods of IPIX micro-motion 1 target in Example 1 of the present invention.

[0046] Figure 3 It is a time-frequency spectrum comparison diagram of sea clutter suppression of different methods of IPIX micro-motion 2 target in Example 1 of the present invention.

[0047] Figure 4 It is a time-frequency spectrum comparison diagram of sea clutter suppression using different methods for CSIR measured data in Example 1 of the present invention.

[0048] Figure 5 It is a comparison diagram of sea clutter suppression SCNR of different images of CSIR measured data according to Example 1 of the present invention.

[0049] Figure 6 This is a comparison diagram of sea clutter suppression SINR of different images of CSIR measured data according to Example 1 of the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0051] Example 1

[0052] Reference Figure 1 , a radar clutter suppression method based on a diffusion model and low-rank matrix decomposition of this embodiment includes:

[0053] Acquire radar signal data;

[0054] Preprocess and fill in the radar signal data;

[0055] The diffusion model is used to simulate the diffusion process of radar signals, and the features of radar signals are extracted to obtain the clutter image of radar signals.

[0056] The clutter image is reduced to the latent space through the autoencoder to obtain the noisy image;

[0057] The noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image;

[0058] The clutter suppression of radar signal data is evaluated by loss function, SCNR and SINR.

[0059] The specific steps include:

[0060] S1. Radar data preprocessing. Radar data preprocessing is a key step in radar system data analysis, which aims to improve data quality, reduce noise, and prepare data for further processing and analysis.

[0061] The purpose of filtering is to remove noise and unnecessary frequency components in the signal and retain the useful signal part. The basic idea of ​​mean filtering is to smooth the signal by replacing each point in the signal with the average value of the points in its neighborhood. This can reduce the impact of high-frequency noise and enhance the smoothness of the signal. Input signal x n , output signal y n , the expression of mean filtering with a window size of 2M+1 is:

[0062]

[0063] y n is the signal value after filtering, x n+k is the value of the original signal in the window, and M is the half-width of the window, which determines the smoothness of the filter. A larger window can provide better smoothing, but may cause loss of signal details; a smaller window can retain more signal details, but the noise removal effect is poor.

[0064] S2, radar value filling. At the boundary of the signal, since the window exceeds the signal range, the following methods can be used for processing; suppose the input signal is x n , whose length is N. For a point x outside the signal range n , mirror filling can be performed using the following formula:

[0065]

[0066] x -n is the value of the signal at position -n, applicable when n<0. 2N-n-2 It is the value of the signal at position 2N-n-2, which is applicable to the case where n≥N. By mirror padding, the signal boundary can be effectively processed to ensure that the mean calculation of the boundary points will not be affected by the missing data. Mirror padding is often used for edge-preserving filtering in image processing and smoothing filtering in signal processing.

[0067] S3. Radar clutter characteristics simulation: The SD diffusion model can simulate the random diffusion behavior of clutter signals in space, which helps to understand the spatial distribution characteristics and change laws of clutter. The diffusion of clutter signals is assumed to be a random process, which is described by the stochastic differential equation (SDE).

[0068] The diffusion of clutter signals occurs in a defined spatial domain, which may be two-dimensional or three-dimensional. Assume that C(t,x) represents the intensity of the clutter signal at time t and position x. The SD diffusion model can be described by the following stochastic differential equation:

[0069]

[0070] D is the diffusion coefficient, which describes the speed of diffusion of clutter signals. ▽2 is the Laplace operator, which represents the diffusion effect in space. σ is the noise intensity, which represents the intensity of random perturbations. dW(t,x) is a Wiener process or Brownian motion, which represents random perturbations.

[0071] The value of the clutter signal at the boundary is fixed to a certain constant.

[0072]

[0073] Simulate the diffusion process: Set C(0,x) according to the initial conditions, use numerical methods (such as finite difference method, finite element method) to gradually solve the stochastic differential equation, and update the value of C(t,x) at each time step Δt. In each time step, according to the noise intensity The nanoprocess dW(t,x) generates random disturbances to simulate the random diffusion behavior of clutter signals. In numerical implementation, the discrete form of the SD diffusion model is usually used:

[0074]

[0075] Δt is the time step, η(t,x) is a standard normally distributed random variable, which represents the random disturbance in the discrete time step. Through the above SD diffusion model, the random diffusion behavior of clutter signals in space can be simulated, which helps to understand its spatial distribution characteristics and change laws. This is of great significance for clutter suppression and analysis in the fields of radar signal processing, image processing, etc.

[0076] S4. Feature extraction: By extracting the features of clutter signals through the SD diffusion model, it is possible to better distinguish them from target signals, thereby improving the accuracy of clutter suppression. By extracting the features of clutter signals through the SD diffusion model, the diffusion coefficient, noise intensity, and spatial distribution characteristics involved in the model can be used to distinguish clutter signals from target signals, thereby improving the accuracy of clutter suppression.

[0077]

[0078] D is the diffusion coefficient. 2 is the Laplace operator. σ is the noise intensity. dW(t,x) is the Wiener process (or Brownian motion). For a given signal C(t,x) to calculate its diffusion term and random term: extract the diffusion feature by calculating the Laplace operator of the signal.

[0079] Diffusion term: by calculating the Laplacian of the signal Extract diffusion features. Extract random features by estimating the noise intensity σ and the contribution of the Wiener process dW(t,x).

[0080] The clutter signal distribution C(t,x) at a series of time points is obtained by simulating the SD diffusion model. Features are extracted from these distributions. Autocorrelation function The autocorrelation of the signal at each position over time.

[0081]

[0082] By extracting the features of clutter signals through the SD diffusion model, we can better understand the spatial distribution characteristics and change patterns of clutter. The feature extraction steps include signal simulation, feature calculation and classification recognition, which helps to improve the accuracy of clutter suppression.

[0083] S5. Use a pre-trained universal variational autoencoder. First use the encoder to reduce the dimension of the input image to the latent space (usually downsampling by a factor of 4-16).

[0084] S6. Latent space representation of radar image. Convert the radar sea clutter image x0 to the latent space feature representation, denoted as z0.

[0085] The forward diffusion process of the diffusion model can be described as:

[0086]

[0087] zt is the feature representation at step t. βt is the noise weight at step t. εt is the noise sampled from a standard normal distribution.

[0088] Starting from t=0t = 0t=0, gradually increase the noise until pure noise is generated. The specific steps are as follows:

[0089] Initialization: z0=x0 (original radar image) Set the total number of steps T and the noise weight sequence For each time step t from 0 to T-1, the noise is sampled Calculate z t+1 :

[0090]

[0091] Record the noise image zt generated at each step as GT.

[0092] S7. Reverse diffusion (reverse sampling) The goal of reverse diffusion is to generate an image or a new image similar to the original image by gradually reducing the noise starting from pure noise. The process of reverse diffusion can be described as:

[0093]

[0094] zt is the feature representation at step t. βt is the noise weight at step t. εt is the noise sampled from a standard normal distribution. A noise prediction model is usually used. To estimate the noise εt to reduce the impact of noise.

[0095] From pure noise z T At the beginning, through multiple iterations, the noise is gradually reduced to generate the target image.

[0096] Sample pure noise from a standard normal distribution Set the total number of steps T and the noise weight sequence .

[0097] Backward diffusion: For each time step t from T-1 to 0, predict the noise ,calculate Record the image zt generated at each step. Through this reverse diffusion process, an image similar to the original image or a new image can be generated from pure noise. Therefore, this process can be used to migrate the sea clutter image to the sea clutter-free image and finally generate the expected clear clutter-free radar image.

[0098] S8. Define the loss function used to train the noise prediction model. A common choice is the mean squared error (MSE), which measures the difference between the predicted noise and the true noise.

[0099]

[0100] θ represents the model parameters. z0 is the original radar sea clutter image. εt is the real noise added during the forward diffusion process. t is the time step. zt is the image representation at time step t. εθ(zt,t) is the output of the noise prediction model, i.e., the noise predicted from zt at time step t.

[0101] S9. Clutter suppression of radar images based on the above diffusion process. The image generated after clutter suppression using the diffusion model. Compared with the image containing clutter, the image after clutter suppression should be clearer, the target objects and features are easier to identify, and the influence of clutter is significantly reduced. The suppression effect of the generated radar image is evaluated by comparing SCNR and SINR.

[0102] S10. Evaluate the clutter suppression of radar image data by SCNR. SCNR refers to the ratio of target signal power to clutter and noise power. The higher the SCNR, the more obvious the target signal is relative to clutter and noise.

[0103]

[0104] First, the target area and background area are extracted from the image to calculate the power of the signal and interference (clutter and noise). The signal power of the target area and the interference power of the background area are calculated.

[0105] S11. Evaluate the clutter suppression of radar image data by SINR. SINR refers to the ratio of target signal power to interference and noise power. The higher the SINR, the more obvious the target signal is relative to interference and noise. The signal to clutter and noise ratio of a cluttered image is usually significantly reduced due to the introduction of clutter and noise. By calculating and comparing SCNR and SINR, the effect of radar image clutter suppression can be quantified to verify whether the image quality after suppression is significantly improved.

[0106] Figure 1 It is the process framework diagram of the present invention. Figure 1It can be seen that the proposed radar clutter suppression method combining diffusion model and low-rank matrix decomposition technology preprocesses radar images to ensure data standardization and stability. The diffusion model simulates the noise diffusion process by gradually adding noise until a pure noise image is generated. In the forward process, the diffusion model is used to gradually transform the image into a noise image. The noise data of each step is recorded for subsequent training. The noise prediction model is trained to predict and remove the noise generated at each step. Starting from the pure noise image, the trained model is used to gradually denoise and restore the image. The signal to clutter and noise ratio (SCNR) and signal to interference and noise ratio (SINR) of the denoised image are calculated and compared. The denoising effect is evaluated to ensure the accuracy and reliability of target detection and feature recognition. Figure 2 This is a comparison of the time-frequency spectrum of sea clutter suppression using different methods for IPIX micro-motion 1 target. Each sub-figure shows the time-frequency spectrum under different processing methods, so as to compare the effects of various sea clutter suppression techniques. Figure 3 It is the comparison of time-frequency spectra of sea clutter suppression by different methods of IPIX micro-motion 2 target. By comparing the time-frequency spectra processed by different methods, the suppression effect of each method on sea clutter can be evaluated. Figure 4 The time-frequency spectrum comparison of different sea clutter suppression methods using CSIR measured data is shown in Figure 2.) The measured data usually include observation data from actual radar systems, which may be more complex and have characteristics in practical applications. Figure 4 The comparison can help evaluate the effects of different clutter suppression methods on real data. Figure 5 and Figure 6 They are comparative experimental diagrams of the method proposed in the present invention and the classic deep learning method in terms of SCNR and SINR indicators. It can be seen from the figure that the method proposed in the present invention is generally superior to the classic deep learning method in terms of SCNR indicator. This shows that the method of the present invention performs better in improving the ratio of the signal to clutter and noise, and can more effectively enhance the clarity of the target signal. It can be seen from the figure that the method proposed in the present invention is also superior to the classic deep learning method in terms of SINR indicator. This shows that the method of the present invention can more effectively improve the ratio of the signal to interference and noise, thereby improving the quality and recognizability of the target signal.

[0107] In summary, the radar clutter suppression method proposed in the present invention, which combines the diffusion model and the low-rank matrix decomposition technology, has great improvements in both image generation and radar image processing.

[0108] The radar clutter suppression method combining a diffusion model and a low-rank matrix decomposition technology of the present invention can effectively improve the quality of radar images through a radar clutter suppression method combining a diffusion model and a low-rank matrix decomposition technology. The method first uses a low-rank matrix decomposition technology to decompose the radar image into a low-rank component (representing the target signal) and a sparse component (representing clutter and noise). Then, the diffusion model is applied to process the decomposed sparse components to simulate and eliminate clutter and noise. The diffusion model is iterated through multiple steps, and noise is gradually added in each step until a pure noise image is generated. In the reverse diffusion process, the noise is gradually removed by a trained noise prediction model to restore a clearer image. Combining the prior knowledge of low-rank matrix decomposition and the powerful processing power of the diffusion model, the method can effectively separate and suppress clutter and noise in radar images. Finally, by calculating and comparing the signal to clutter and noise ratio (SCNR) and the signal to interference and noise ratio (SINR), the effect of clutter suppression can be verified to ensure that the generated image has higher reliability and accuracy in target detection and feature recognition.

[0109] Example 2

[0110] This embodiment provides a radar clutter suppression system based on a diffusion model and low-rank matrix decomposition, including:

[0111] A data acquisition module is configured to acquire radar signal data;

[0112] The preprocessing module is configured to preprocess and fill in the radar signal data;

[0113] The feature extraction module is configured to simulate the diffusion process of the radar signal using a diffusion model and extract features of the radar signal to obtain a clutter image of the radar signal;

[0114] The noise module is configured to reduce the dimension of the noise image to the latent space through the autoencoder to obtain the noise image;

[0115] The diffusion module is configured to perform forward diffusion and reverse diffusion on the noise image in sequence to generate a clutter-free radar image;

[0116] The evaluation module is configured to evaluate the clutter suppression of the radar signal data through the loss function, SCNR and SINR.

[0117] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method described.

[0118] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; and the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the described method.

[0119] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A radar clutter suppression method based on a diffusion model and low-rank matrix decomposition, characterized in that: include: Acquire radar signal data; Preprocess and fill in the radar signal data; The diffusion model is used to simulate the diffusion process of radar signals, and the features of radar signals are extracted to obtain the clutter image of radar signals. The clutter image is reduced to the latent space through the autoencoder to obtain the noisy image; The noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image; The clutter suppression of radar signal data is evaluated by loss function, SCNR and SINR; The preprocessing and value filling of the radar signal data includes filtering the radar data to remove noise and unnecessary frequency components in the radar signal, wherein the mean filter is used to smooth the signal, which is expressed as: ， in, is the signal value after filtering, is the value of the original signal in the window, M is the half-width of the window, which determines the smoothness of the filter, and the input signal , output signal , the window size is 2M+1; The preprocessing and value filling of the radar signal data also includes mirror filling of the radar signal beyond the window range at the signal boundary, which is expressed as: ， The input signal is , whose length is N, is the value of the signal at position -n, applicable to the case where n < 0, is the value of the signal at position 2N-n-2, applicable to the case where n ≥ N; The diffusion model is used to simulate the diffusion process of the radar signal, including the diffusion model being described by the following stochastic differential equation: ， D is the diffusion coefficient, which describes the speed at which the clutter signal diffuses. is the Laplace operator, which represents the diffusion effect in space, is the noise intensity, which indicates the intensity of random disturbances, and dW(t,x) is a Wiener process or Brownian motion, which indicates random disturbances and simulates the diffusion process: according to the initial conditions, C ( 0,x ), using numerical methods to gradually solve the stochastic differential equation, updating the value of C(t,x) at each time step Δt. In each time step, according to the noise intensity The Wiener process dW(t,x) generates random disturbances to simulate the random diffusion behavior of clutter signals. The diffusion model is used to simulate the random diffusion behavior of clutter signals in radar signals in space. The discrete diffusion model is used, which is expressed as: ， Where Δt is the time step, η(t,x) is a standard normally distributed random variable representing the random perturbation in the discrete time step, and the noise intensity , represents the intensity of random disturbance; D is the diffusion coefficient, C(t,x) represents the intensity of the clutter signal at time t and position x, is the Laplace operator, representing the diffusion effect in space; The feature extraction of the radar signal includes distinguishing the clutter signal and the target signal in the radar signal based on the diffusion coefficient, the noise intensity and the spatial distribution characteristics, extracting the diffusion feature by calculating the Laplace operator of the clutter signal, extracting the random feature by estimating the noise intensity and the contribution of the Wiener process, and finally obtaining the autocorrelation of the radar signal over time by simulating the distribution of the clutter signal; The noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image, including converting the noise image x0 into a latent space feature representation, denoted as z0. The forward diffusion process of the diffusion model is: ， in, is the feature representation of the tth step, is the noise weight at step t, is the noise sampled from the standard normal distribution, starting from t=0, and gradually increasing the noise until pure noise is generated; The noise image is forward diffused and reverse diffused in sequence to generate a clutter-free radar image, and the reverse diffusion of pure noise is also performed, which is expressed as: ， in, is the feature representation of the tth step, is the noise weight at step t, is the noise sampled from a standard normal distribution, using a noise prediction model Estimated noise To reduce the impact of noise, from pure noise At the beginning, through multiple steps of iteration, the noise is gradually reduced to generate the target image; The suppression effect of the generated radar image is evaluated by comparing SCNR and SINR.

2. A radar clutter suppression system based on a diffusion model and low-rank matrix decomposition, executing the method according to claim 1, characterized in that: include: A data acquisition module is configured to acquire radar signal data; The preprocessing module is configured to preprocess and fill in the radar signal data; The feature extraction module is configured to simulate the diffusion process of the radar signal using a diffusion model and extract features of the radar signal to obtain a clutter image of the radar signal; The noise module is configured to reduce the dimension of the noise image to the latent space through the autoencoder to obtain the noise image; The diffusion module is configured to perform forward diffusion and reverse diffusion on the noise image in sequence to generate a clutter-free radar image; The evaluation module is configured to evaluate the clutter suppression of the radar signal data through the loss function, SCNR and SINR.

3. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing a radar clutter suppression method based on a diffusion model and low-rank matrix decomposition as claimed in claim 1.

4. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the radar clutter suppression method based on a diffusion model and low-rank matrix decomposition as claimed in claim 1.

Images