A radar image domain interference detection and suppression method

Abstract

The present invention proposes a radar image domain interference detection and suppression method, comprising: dividing a complete SAR single-view complex image into blocks; preprocessing each block, including: using CFAR to detect and remove strong scattering points; performing interference detection on each preprocessed block; and using a block subspace filter BSF to suppress interference for blocks where interference is detected. The present invention can effectively distinguish interference and non-interference areas, thereby realizing adaptive detection and suppression of interference and improving radar image quality.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar signal processing, and in particular relates to a radar image domain interference detection and suppression method. Background Art

[0002] Radar imaging techniques, such as Synthetic Aperture Radar (SAR), use electromagnetic waves to obtain radar images of the Earth's surface and are the basis for many remote sensing applications. However, the quality of radar data can be affected by interference from a variety of radio signals, including ground-based and space-based sources. Ground-based sources include commercial and military radio transmission systems, such as FM / TV broadcasts, radar systems, etc., which transmit signals that may interfere with imaging radar systems operating in similar frequency bands. Similarly, space sources such as communication satellites or other radar satellites may also cause interference if they operate in close proximity or share the same frequency band. Military jammers can also cause interference to imaging radars, which is a form of deliberate interference that interferes with radar systems by emitting strong signals to suppress radar signals. This may be caused by military or security systems designed to block or degrade the performance of imaging radar systems. The effects of interference on image quality are significant, including reduced signal-to-noise ratio, dynamic range distortion, image blurring, amplitude and phase distortion, and reduced spatial resolution. In the field of remote sensing, interference detection and suppression in imaging radar data has become a key research topic. This paper focuses on the detection stage. Many methods have been proposed for interference detection in imaging radar data, and many of these works use the processing pipeline of traditional signal processing methods to perform interference detection. These methods first extract interference features in certain domains (e.g., frequency domain, multi-channel domain, eigenvalue domain) through tools such as spectral analysis and eigenvalue decomposition, and then use these features for interference detection through statistical tests, empirical thresholds, or clarification methods such as k-means clustering. With the progress of efficient deep neural networks (e.g., U-net, SSD, ResNet) in the field of computer vision, some studies have used neural networks to learn interference features from data and automatically detect interference. Although deep learning-based methods have shown promise in imaging radar interference detection, the training of neural networks requires a large amount of labeled data, and the generalization ability of these networks to different imaging radar data sources may be problematic. Summary of the invention

[0003] The present invention proposes a radar image domain interference detection and suppression method, which can effectively identify and remove interference artifacts in radar images and improve image quality and usability.

[0004] The technical solution to achieve the purpose of the present invention is: a radar image domain interference detection and suppression method, comprising:

[0005] Step 1: Segment the complete SAR single-view complex image into blocks;

[0006] Step 2: Preprocess each block, including: using CFAR to detect strong scattering points and remove them;

[0007] Step 3: Perform interference detection on each preprocessed block;

[0008] Step 4: For the blocks where interference is detected, a block subspace filter (BSF) is used to suppress interference.

[0009] Preferably, the size of the SAR single-view complex image is N a ×N r , the number of blocks divided into is: floor(N a / L)×floor(N r / L), L is the side length of the block.

[0010] Preferably, the specific method for performing interference detection on each preprocessed block is:

[0011] Construct the covariance matrix of the preprocessed data block;

[0012] The maximum eigenvalue of the covariance matrix is ​​determined, and the maximum eigenvalue is compared with a threshold value. If the maximum eigenvalue of the covariance matrix exceeds the threshold value, it is determined that interference exists in the current data block.

[0013] Preferably, the covariance matrix of the preprocessed data block is constructed as: is the preprocessed image block.

[0014] Preferably, the threshold is ξλ th , where ξ≥1 is a multiplicative hyperparameter, express The inverse function of 2 is the variance, a and b are scaling and translation parameters respectively, P fa is the false alarm rate, is the cumulative distribution function:

[0015]

[0016] where Γ(k) is the gamma function, It is an incomplete gamma function, where the parameters are k=79.6595, θ=0.101037, and α=9.81961.

[0017] Preferably, for a block where interference is detected, a specific method of using a block subspace filter BSF to suppress interference is:

[0018] Perform eigenvalue decomposition on the covariance matrix, specifically:

[0019] R=UΣU H

[0020] Among them, U is the decomposition matrix, the column vector of U is the eigenvector of the covariance matrix R, Σ=diag(λ1,…,λ N ),λ1,…,λ N are the eigenvalues ​​of the covariance matrix R in descending order;

[0021] Use the K eigenvectors corresponding to the K largest eigenvalues ​​to construct the interference subspace U 1:K ;

[0022] The image block S masked Projecting into the interference subspace to reconstruct the interference component Specifically:

[0023]

[0024] By subtracting the reconstructed interference component from the original image block, an estimate of the clean image is obtained, i.e.

[0025]

[0026] Where S is the original image block.

[0027] Compared with the prior art, the present invention has the following significant advantages: the present invention can effectively distinguish interference and non-interference areas, thereby realizing adaptive detection and suppression of interference and improving the quality of radar images.

[0028] The present invention is described in further detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of image segmentation.

[0030] Figure 2 Comparison between the maximum eigenvalue (red dot) and the threshold (blue line) of 100 simulated data blocks without interference.

[0031] Figure 3 An image with interference.

[0032] Figure 4 The block interference detection results are shown in red, indicating interference.

[0033] Figure 5 This is the result after adaptive interference suppression. DETAILED DESCRIPTION

[0034] The present invention, namely a radar image domain interference detection and suppression method, is further described below with reference to the accompanying drawings and examples.

[0035] A radar image domain interference detection and suppression method is proposed. The maximum eigenvalue of the imaging radar image data block is analyzed and compared with a preset threshold to detect whether there is interference in the image. The method includes image segmentation, calculating the covariance matrix of each data block, extracting the maximum eigenvalue, and determining the threshold using the constant false alarm rate (CFAR) criterion. Then the maximum eigenvalue of each block is compared with the threshold to determine whether the data block has interference. For the blocks with interference, the block subspace algorithm is used to suppress interference. The specific steps are as follows:

[0036] Step 1: Complete N a ×N r SAR single-view complex (SLC) image segmentation into floor (N a / L)×floor(N r / L) blocks of size L×L pixels. Here, the mathematical symbol floor(x) represents the floor function, which rounds x down to the nearest integer less than or equal to x. Figure 1 shown.

[0037] Step 2: Perform preprocessing on each block. Use CFAR to detect strong scattering points and remove them. The preprocessed image block is denoted as S masked .

[0038] Step 3: For each preprocessed block, perform interference detection. First, construct the covariance matrix of the current data block:

[0039]

[0040] Then, the maximum eigenvalue λ1 of the current block covariance matrix is ​​calculated.

[0041] Finally, λ1 is compared with a certain threshold. If the maximum eigenvalue of the covariance matrix exceeds the threshold, it is determined that there is interference in the current data block.

[0042] In a further embodiment, the threshold selection process is as follows:

[0043] Detecting the presence of interference in radar images can be conceptualized as a binary hypothesis testing process, where two scenarios are considered: 1) H0: The image is homogeneous and contains only background clutter and noise, with no interference artifacts. 2) H1: The image has interference artifacts superimposed on a homogeneous background. Under the homogeneity assumption (H0), the eigenvalues ​​of the image should ideally exhibit stable behavior and be constrained to a predictable range. In order to perform detection under the CFAR criterion, the distribution of the maximum eigenvalue needs to be obtained, based on which the maximum eigenvalue threshold at a given false alarm rate can be selected. Eigenvalues ​​exceeding this threshold then indicate the presence of interference under the CFAR criterion.

[0044] Assume that each image block is an N×M matrix S masked , and each element in the image block is an independent and identically distributed zero-mean complex Gaussian random variable with variance σ 2 (i.e. the image is uniform). It is worth noting that in practice radar image pixels may not be statistically independent, however, for theoretical simplicity, the correlation between pixels will be ignored. Covariance matrix is an uncorrelated complex Wishart matrix. In the case of large dimensions N×M, The maximum eigenvalue of can be approximated by the second type Tracy-Widom distribution (TW2). Specifically, the maximum eigenvalue λ1 is translated and scaled to obtain the variable x, which obeys the second type Tracy-Widom distribution:

[0045]

[0046] λ1 is the maximum eigenvalue of the covariance matrix of the current data block; a and b are scaling and translation parameters respectively;

[0047] In a further embodiment, the scaling and translation parameters a and b are respectively specifically:

[0048]

[0049]

[0050] The cumulative distribution function of a random variable x that obeys TW2 distribution can be approximately expressed as:

[0051]

[0052] where Γ(k) is the gamma function, It is an incomplete gamma function, where the parameters are k=79.6595, θ=0.101037, and α=9.81961.

[0053] If the largest eigenvalue of the covariance matrix exceeds a sufficiently large threshold λ th , then it can be judged that the image contains interference. The false alarm rate (there is actually no interference, but the maximum eigenvalue exceeds the threshold, resulting in the false detection of interference) can be modeled as follows

[0054]

[0055] Based on this equation, the threshold λ is derived th The relationship with the false alarm rate is as follows

[0056]

[0057] In the formula, Indicates F TW2 The inverse function of (x).

[0058] Figure 2 The maximum eigenvalue and threshold calculated from 100 simulation data without interference are given.

[0059] The above discussion assumes that the observed scene is uniform and its image pixels satisfy an independent and identically distributed Gaussian distribution. However, ideal uniform scenes rarely exist in real scenes, so actual radar images will deviate from the Gaussian distribution to varying degrees. This causes the actual maximum eigenvalue to be larger than that under ideal conditions, which may lead to an increase in the false alarm rate. To alleviate this problem, an additional multiplicative hyperparameter ξ ≥ 1 is introduced and ξλ is used th as the actual detection threshold.

[0060] Step 4: For the blocks where interference is detected, interference suppression is performed using the block subspace filter BSF. Specifically, the filter first performs eigenvalue decomposition on the covariance matrix

[0061] R=UΣU H

[0062] where U is a matrix whose columns are the eigenvectors of R, and Σ = diag(λ1,…,λ N ) is a diagonal matrix containing the eigenvalues ​​of R. Here we assume that the eigenvalues ​​are arranged in descending order. Next, the filter constructs the interference subspace using the K eigenvectors corresponding to the K largest eigenvalues, i.e., U 1:K Then, the image block S masked Projecting into the interference subspace to reconstruct the interference component is recorded as

[0063]

[0064] Finally, by subtracting the reconstructed interference component from the original image patch, an estimate of the clean image is obtained, i.e.

[0065]

[0066] Where S is the original image block.

[0067] Figure 3 An example of an image with interference is given. Figure 4 The interference block detection results are given. Figure 5 The results after adaptive interference suppression are given.

Claims

1. A radar image domain interference detection and suppression method, characterized in that: include: Step 1: Segment the complete SAR single-view complex image into blocks; Step 2: Preprocess each block, including: using CFAR to detect strong scattering points and remove them; Step 3: Perform interference detection on each preprocessed block. The specific method is as follows: Construct the covariance matrix of the preprocessed data block, specifically: S masked is the preprocessed image block; Determine the maximum eigenvalue of the covariance matrix and compare it with the threshold. If the maximum eigenvalue of the covariance matrix exceeds the threshold, it is determined that there is interference in the current data block; the threshold is ξλ th , where ξ≥1 is a multiplicative hyperparameter, express The inverse function of 2 is the variance, a and b are scaling and translation parameters respectively, P fa is the false alarm rate, is the cumulative distribution function: where Γ(k) is the gamma function, is the incomplete gamma function, where the parameters are k = 79.6595, θ = 0.101037, α = 9.81961; Step 4: For the blocks where interference is detected, use the block subspace filter BSF to suppress interference. The specific method is as follows: Perform eigenvalue decomposition on the covariance matrix, specifically: R=UΣU H Among them, U is the decomposition matrix, the column vector of U is the eigenvector of the covariance matrix R, Σ=diag(λ1,…,λ N ),λ1,…,λ N are the eigenvalues ​​of the covariance matrix R in descending order; Use the K eigenvectors corresponding to the K largest eigenvalues ​​to construct the interference subspace U 1:K ; The image block S masked Projecting into the interference subspace to reconstruct the interference component Specifically: By subtracting the reconstructed interference component from the original image block, an estimate of the clean image is obtained, i.e. Where S is the original image block.

2. The radar image domain interference detection and suppression method according to claim 1, characterized in that: The size of the SAR single-view complex image is N a ×N r , the number of blocks divided into is: floor(N a / L)×floor(N r / L), L is the side length of the block.

Images