Multi-channel ground moving target detection method based on RPCA-STAP

Abstract

The application discloses a kind of multi-channel ground moving target detection methods based on RPCA-STAP, comprising: according to Markov random field image segmentation algorithm, the SAR image to be detected is handled, and the image after segmentation is obtained;Wherein, the image after segmentation includes the sub-image of different clutter types;The clutter type of each sub-image has consistency;And according to RPCA-STAP algorithm, it is handled, and low rank matrix and sparse matrix are obtained;The low rank matrix and sparse matrix are updated by exchanging direction multiplier method, and the updated low rank matrix and sparse matrix are obtained;And based on this, moving target is detected, and the position of moving target is obtained.The application effectively solves the problem that traditional technology is difficult to distinguish moving target and background clutter due to sample pollution or sample unevenness, thereby leading to the decline of moving target detection performance, significantly improves the accuracy and robustness of target detection;Reduce the complexity of calculation, improve the practicability of algorithm.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of synthetic aperture radar target detection, and particularly relates to a multi-channel ground moving target detection method based on RPCA-STAP. BACKGROUND

[0002] Synthetic aperture radar has all-weather, all-day and high-resolution imaging capabilities, and is widely used in military reconnaissance, ocean monitoring, traffic monitoring, disaster monitoring and other fields. The detection of ground moving targets (Ground Moving Target Indication, GMTI) by a multi-channel synthetic aperture radar (Synthetic Aperture Radar, SAR) system is one of the important research directions of SAR application, and has high military and civilian value.

[0003] When a multi-channel synthetic aperture radar-Ground Moving Target Indication (SAR-GMTI) system works downward, the ground clutter energy seriously interferes with the detection of slow-moving targets. This interference can cause the target signal and the clutter signal to overlap in the range-Doppler domain, and the Signal-to-Clutter-plus-Noise Ratio (SCNR) is significantly reduced, so that it is difficult to effectively distinguish the moving target from the background clutter, and the ground moving target indication performance is affected. Therefore, effectively suppressing clutter and improving the performance of moving target detection has become an important research direction of SAR ground moving target detection technology. SUMMARY

[0004] In order to solve the above problems existing in the prior art, the application provides a multi-channel ground moving target detection method based on RPCA-STAP.

[0005] The technical problem to be solved by the application is solved by the following technical scheme:

[0006] In a first aspect, the application provides a multi-channel ground moving target detection method based on RPCA-STAP, which comprises:

[0007] The SAR image to be detected is processed according to the Markov random field image segmentation algorithm to obtain a segmented image; wherein the segmented image comprises sub-images of different clutter types; the clutter type of each sub-image is consistent;

[0008] The segmented image is processed according to the pre-built RPCA-STAP algorithm to obtain a low-rank matrix and a sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm;

[0009] The low-rank matrix and sparse matrix corresponding to each sub-image are updated by exchanging direction multipliers, resulting in the updated low-rank matrix and sparse matrix of each sub-image.

[0010] Based on the updated low-rank matrix and the updated sparse matrix of each sub-image, the moving targets in the SAR image to be detected are detected to obtain the position of the moving targets.

[0011] Optionally, the RPCA-STAP algorithm is expressed as follows:

[0012]

[0013] Among them, ||L|| * Let γ||S||1 denote the nuclear norm of the low-rank matrix L, γ be the l1 norm of the sparse matrix S, γ be a hyperparameter controlling sparsity, and αw be the nuclear norm of the low-rank matrix L. H ZZ H w is the adaptive weighting term, w is the adaptive weight, Z is the auxiliary variable generated by the low-rank matrix L, Z H Z is the conjugate transpose of Z, ZZ H The autocorrelation matrix of Z is given by α, where α is the hyperparameter of adaptive suppression; st represents the constraint condition in the above equation. The constraints of the Frobenius norm are represented, δ represents the tolerable error size, X represents the segmented image, V(L) represents the conjugate transpose of the low-rank matrix L, and |S i |=|S i-1 | is used to ensure that the magnitude of each column in the sparse matrix S remains consistent, i is the column number of the sparse matrix S, N is the total number of columns in the sparse matrix S, and w(1) = 1 is used to ensure that the first element of the adaptive weight is 1.

[0014] Optionally, updating the low-rank matrix and sparse matrix corresponding to each sub-image using the exchange-direction multiplier method to obtain the updated low-rank matrix and sparse matrix of each sub-image includes:

[0015] An augmented Lagrangian function is constructed based on the RPCA-STAP algorithm, the low-rank matrix corresponding to each sub-image, and the sparse matrix corresponding to each sub-image.

[0016] The augmented Lagrangian function is updated by exchanging direction multipliers to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image.

[0017] Optionally, the augmented Lagrangian function is expressed as follows:

[0018]

[0019] Where μ1 represents the first penalty factor, μ2 represents the second penalty factor, Y1 represents the first Lagrange multiplier, Y2 represents the second Lagrange multiplier, and <, > represent the matrix inner product.<Y1,X-L-S> Denotes the first Lagrange multiplier term.<Y2,V(L)-Z> This represents the second Lagrange multiplier term.

[0020] Optionally, updating the augmented Lagrangian function using the method of exchanging direction multipliers to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image includes:

[0021] By fixing the variables in the augmented Lagrangian function except for the low-rank matrix, the low-rank matrix is ​​updated by minimizing the augmented Lagrangian function, thus obtaining the updated low-rank matrix for each sub-image.

[0022] By fixing the variables in the augmented Lagrangian function except for the sparse matrix, the sparse matrix is ​​updated by minimizing the augmented Lagrangian function, thus obtaining the updated sparse matrix for each sub-image.

[0023] Optionally, the process of updating the low-rank matrix by minimizing the augmented Lagrangian function, while keeping all variables in the augmented Lagrangian function fixed, is represented as follows:

[0024]

[0025] Optionally, the process of updating the sparse matrix by minimizing the augmented Lagrangian function, while fixing all variables except the sparse matrix, is represented as follows:

[0026]

[0027] Secondly, the present invention provides a multi-channel ground moving target detection device based on RPCA-STAP, the device comprising:

[0028] The image segmentation module is used to process the SAR image to be detected according to the Markov random field image segmentation algorithm to obtain the segmented image; wherein, the segmented image includes sub-images with different clutter types; the clutter type of each sub-image is consistent;

[0029] The image processing module is used to process the segmented image according to the pre-built RPCA-STAP algorithm to obtain a low-rank matrix and a sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm;

[0030] The matrix update module is used to update the low-rank matrix and the sparse matrix corresponding to each sub-image by exchanging direction multipliers, so as to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image.

[0031] The target detection module is used to detect moving targets in the SAR image to be detected based on the updated low-rank matrix and the updated sparse matrix of each sub-image, and to obtain the position of the moving targets.

[0032] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0033] In the above technical solution, the Markov random field image segmentation algorithm is used to classify clutter types in SAR images, ensuring that each sub-image satisfies the independent and identically distributed condition. This effectively solves the problem that traditional techniques are unable to distinguish moving targets from background clutter due to sample contamination or unevenness, which leads to a decline in moving target detection performance. The RPCA-STAP algorithm is constructed, and the strong clutter suppression capability of the STAP algorithm is utilized to effectively reduce the false alarm rate of the RPCA algorithm in strong clutter backgrounds, significantly improving the accuracy and robustness of target detection. The alternating direction multiplier method is used to update the low-rank matrix and sparse matrix obtained from the RPCA-STAP algorithm, which can reduce computational complexity and improve the practicality of the algorithm.

[0034] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0035] Figure 1 This is a flowchart of a multi-channel ground moving target detection method based on RPCA-STAP provided in an embodiment of the present invention;

[0036] Figure 2a This is a schematic diagram of a road image with weak energy in an MRF image segmentation result provided by an embodiment of the present invention;

[0037] Figure 2b This is a schematic diagram of a grassland image with weak energy in an MRF image segmentation result provided by an embodiment of the present invention;

[0038] Figure 2cThis is a schematic diagram of a building image with high energy in an MRF image segmentation result provided by an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the detection probability results corresponding to different algorithms provided in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of an experimental scenario provided by an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram of a synthetic aperture radar imaging result provided in an embodiment of the present invention;

[0042] Figure 6a This is a schematic diagram of the target detection result of an RPCA algorithm provided in an embodiment of the present invention;

[0043] Figure 6b This is a schematic diagram of the target detection result of the STAP algorithm provided in an embodiment of the present invention;

[0044] Figure 6c This is a schematic diagram of the target detection result of a DPCA algorithm provided in an embodiment of the present invention;

[0045] Figure 6d This is a schematic diagram of the target detection result of an ATI-GoDec algorithm provided in an embodiment of the present invention;

[0046] Figure 6e This is a schematic diagram of the target detection result of the RPCA-STAP algorithm provided in an embodiment of the present invention;

[0047] Figure 7 This is a schematic diagram of the relocation result of a moving target provided in an embodiment of the present invention;

[0048] Figure 8 This is a block diagram of a multi-channel ground moving target detection device based on RPCA-STAP provided in an embodiment of the present invention. Detailed Implementation

[0049] To facilitate understanding of the present invention, a brief description of the prior art and the inventive concept of the present invention will be provided first.

[0050] Traditional signal processing algorithms include: Displaced Phase Center Antenna (DPCA) technology: DPCA eliminates stationary clutter by subtracting the signals from the two receiving channels, retaining only the moving target signal. Its advantages are simplicity and good real-time performance, but it has a high minimum detectable speed, making it difficult to detect slow-moving targets. Along-track Interferometry (ATI) technology: ATI utilizes the phase difference between different channels to detect moving targets. Space-Time Adaptive Processing-based Along-Track Interferometry (SP-ATI) algorithm achieves clutter suppression by constructing a clutter subspace, but lacks adaptability to environmental changes, resulting in decreased detection performance in complex scenarios. Space-Time Adaptive Processing (STAP): STAP suppresses clutter by constructing a two-dimensional filter that combines spatial and temporal degrees of freedom, effectively improving the detection probability of moving targets and is widely used in range-Doppler radar systems. To address the performance degradation caused by target motion during long accumulation times, imaging STAP and a three-dimensional adaptive spatial-temporal processing algorithm are proposed. Extended Displaced Phase Center Antenna (EDPCA): EDPCA extends DPCA technology to any channel and achieves optimal Imaging Space-Time Adaptive Processing (ISTAP) performance under DPCA applicable conditions. By adjusting the target steering vector, a clutter suppression method without parameter search is proposed, significantly improving the system's real-time performance.

[0051] Low-rank matrix recovery algorithms include Robust Principal Component Analysis (RPCA): RPCA models clutter in multi-channel SAR images as a low-rank matrix and moving targets as a sparse matrix, using matrix factorization to detect moving targets. RPCA does not require parameter search and has good practical application value. However, in real-world strong clutter environments, the inevitable leakage of clutter energy into the sparse matrix leads to a significant increase in false targets and a high false alarm rate, severely impacting the accuracy of target detection.

[0052] Improved RPCA algorithms include the ATI-GoDec algorithm, which combines ATI technology with gradient optimization for dimensionality reduction (GoDec) to reduce false alarm rates caused by improper sparse parameter settings. Furthermore, some researchers have extended the RPCA algorithm to propose the Fully Independent Robust Principal Component Analysis (FI-RPCA) algorithm for moving target detection, and Non-local Self-Similarity Robust Principal Component Analysis (NSS-RPCA). However, most of these methods simply cascade the RPCA and STAP algorithms, failing to establish a unified optimization algorithm and thus failing to fully utilize the information in the low-rank matrix.

[0053] To address the aforementioned technical problems, this invention proposes a multi-channel ground moving target detection method based on RPCA-STAP.

[0054] Figure 1 This is a flowchart of a multi-channel ground moving target detection method based on RPCA-STAP provided in an embodiment of the present invention, as follows: Figure 1 As shown, the method includes the following steps:

[0055] S101. The SAR image to be detected is processed according to the Markov random field image segmentation algorithm to obtain the segmented image; wherein, the segmented image includes sub-images of different clutter types; the clutter type of each sub-image is consistent.

[0056] Understandably, in order to meet the STAP algorithm's requirement for independently and identically distributed (IID) training samples and ensure that clutter in each sub-region has similar statistical characteristics, thereby improving the clutter suppression effect of the STAP algorithm, the Markov Random Field (MRF) image segmentation algorithm can be used to classify SAR images and classify clutter in different regions into different categories (such as cities, grasslands, roads, etc.).

[0057] Specifically, Figure 2a This is a schematic diagram of a road image with weak energy in an MRF image segmentation result provided by an embodiment of the present invention. Figure 2bThis is a schematic diagram of a grassland image with weak energy in an MRF image segmentation result provided by an embodiment of the present invention. Figure 2c This is a schematic diagram of a building image with high energy in an MRF image segmentation result provided by an embodiment of the present invention. Figure 2c It may contain dense scatterers, requiring targeted adjustment of the STAP parameter; Figure 2a This may contain slow-moving targets, requiring the preservation of sparse components. Block processing avoids the impact of globally non-uniform samples on STAP performance. The Markov random field image segmentation algorithm aims to classify and segment images based on local information. SAR images are divided into independent blocks of different clutter types (such as urban areas, grasslands, roads, etc.). This process fully utilizes the spatial context information of the SAR image, ensuring that clutter within the same block has similar statistical characteristics, thus satisfying the STAP algorithm's IID assumption for training samples.

[0058] The Markov random field image segmentation algorithm maximizes the posterior probability using Bayes' theorem. It takes the SAR image to be detected, acquired by a synthetic aperture radar (SAR) system, as input and classifies pixels into their optimal categories. The output classification label is the clutter category corresponding to each pixel, and the segmentation results visually demonstrate the distribution of different regions in the SAR image to be detected. Specifically, for sub-image I, the posterior probability of pixel category w1 is calculated as follows:

[0059]

[0060] Wherein, P(w1) represents the prior probability of class w1, which is the probability that a pixel belongs to class w1 without observing an image; P(I) represents the marginal probability of sub-image I, which serves as a normalization factor to ensure that the sum of the probabilities of all classes is 1; P(w1|I) represents the posterior probability of a pixel belonging to class w1 given sub-image I; and P(I|w1) represents the probability of observing sub-image I given class w1. P(I|w1) and P(w1) are usually modeled using MRF.

[0061] S102. The segmented image is processed according to the pre-constructed RPCA-STAP algorithm to obtain the low-rank matrix and sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm.

[0062] Understandably, in order to combine the low-rank matrix factorization capability of the RPCA algorithm with the strong clutter suppression advantage of the STAP algorithm, an RPCA-STAP algorithm is constructed. This algorithm performs RPCA-STAP joint optimization in each sub-image, thereby meeting the requirements of STAP for IID training samples and suppressing the strong clutter energy leakage problem of RPCA.

[0063] The RPCA-STAP algorithm is represented as follows:

[0064]

[0065] Among them, ||L|| * Let γ||S||1 denote the nuclear norm of the low-rank matrix L, γ be the l1 norm of the sparse matrix S, γ be a hyperparameter controlling sparsity, and αw be the nuclear norm of the low-rank matrix L. H ZZ H w is the adaptive weighting term, w is the adaptive weight, Z is the auxiliary variable generated by the low-rank matrix L, Z H Z is the conjugate transpose of Z, ZZ H The autocorrelation matrix of Z is given by α, where α is the hyperparameter of adaptive suppression; st represents the constraint condition in the above equation. The constraints of the Frobenius norm are represented, δ represents the tolerable error size, X represents the segmented image, V(L) represents the conjugate transpose of the low-rank matrix L, and |S i |=|S i-1 | is used to ensure that the magnitude of each column in the sparse matrix S remains consistent, i is the column number of the sparse matrix S, N is the total number of columns in the sparse matrix S, and w(1) = 1 is used to ensure that the first element of the adaptive weight is 1.

[0066] S103. Update the low-rank matrix and sparse matrix corresponding to each sub-image by exchanging direction multipliers, and obtain the updated low-rank matrix and sparse matrix of each sub-image.

[0067] Optionally, S103 may include:

[0068] An augmented Lagrangian function is constructed based on the RPCA-STAP algorithm, the low-rank matrix corresponding to each sub-image, and the sparse matrix corresponding to each sub-image.

[0069] The augmented Lagrangian function is updated by exchanging direction multipliers, resulting in the updated low-rank matrix and sparse matrix for each sub-image.

[0070] Understandably, the augmented Lagrange function can be expressed as follows:

[0071]

[0072] Where μ1 represents the first penalty factor, μ2 represents the second penalty factor, Y1 represents the first Lagrange multiplier, Y2 represents the second Lagrange multiplier, and <, > represent the matrix inner product.<Y1,X-L-S> Denotes the first Lagrange multiplier term.<Y2,V(L)-Z> This represents the second Lagrange multiplier term.

[0073] Optionally, the augmented Lagrangian function is updated using the method of exchanging direction multipliers to obtain the updated low-rank matrix and the updated sparse matrix for each sub-image, including:

[0074] By fixing the variables in the augmented Lagrangian function except for the low-rank matrix, the low-rank matrix is ​​updated by minimizing the augmented Lagrangian function, thus obtaining the updated low-rank matrix for each sub-image.

[0075] By fixing the variables in the augmented Lagrangian function except for the sparse matrix, the sparse matrix is ​​updated by minimizing the augmented Lagrangian function, resulting in the updated sparse matrix for each sub-image.

[0076] Understandably, the process of updating the low-rank matrix by minimizing the augmented Lagrangian function, while keeping all variables in the augmented Lagrangian function fixed, can be represented as follows:

[0077]

[0078] Solve the above equation using a singular value thresholding operation:

[0079]

[0080] Where U, [Ξ], and V are the results of singular value decomposition;

[0081]

[0082] With the variables in the augmented Lagrangian function fixed except for the sparse matrix, the process of updating the sparse matrix by minimizing the augmented Lagrangian function is represented as follows:

[0083]

[0084] The closed-form solution is shown below:

[0085]

[0086] Among them, S j，i Y represents the element in the i-th column and j-th row of the sparse matrix S. 1，j，i Let Y represent the element in the i-th column and j-th row of the Lagrange multiplier matrix Y1.

[0087] In addition, other parameters in the augmented Lagrangian function can be updated, while keeping the remaining variables fixed and treating Z as a variable, by minimizing the following objective function to update Z:

[0088]

[0089] Directly solving the closed-form solution yields the following:

[0090]

[0091] With the remaining variables fixed and w treated as a variable, update w by minimizing the following objective function:

[0092]

[0093] R z Represents the sample covariance matrix;

[0094] The closed-form solution calculated using the above formula is expressed as follows:

[0095]

[0096] Where a(v) r ) indicates based on radial velocity v r The target direction vector, in order to improve the calculation speed, makes a(v r = [1, 0, ..., 0] T To avoid slowing down the search for the target;

[0097] With the remaining variables fixed and the Lagrange multipliers Y1 and Y2 and the penalty factors μ1 and μ2 treated as variables, the update is performed by minimizing the following objective function:

[0098]

[0099] Where ρ is a constant used to accelerate the algorithm, and μ 1，max and μ 2，max This represents the maximum value of the penalty factor.

[0100] The algorithm stops when the maximum number of iterations is reached or when the following requirement is met.

[0101]

[0102] S104. Based on the updated low-rank matrix and the updated sparse matrix of each sub-image, the moving targets in the SAR image to be detected are detected to obtain the position of the moving targets.

[0103] Understandably, by using the estimated sparse matrix information to relocate the target position based on the optimized low-rank matrix and sparse matrix, combined with the phase difference in multi-channel SAR signals, the accuracy of target positioning can be improved, ensuring that the detected target position is consistent with the actual position, and verifying the effectiveness of target detection.

[0104] In one implementation, prior to S101, the method may further include the step of acquiring the SAR image to be detected: modeling the received signal of the multi-channel synthetic aperture radar data as a superposition of low-rank background clutter and sparse moving targets to obtain the SAR image to be detected.

[0105] In one implementation, Figure 3 This is a schematic diagram illustrating the detection probability results corresponding to different algorithms provided in an embodiment of the present invention. Figure 3 This figure shows a comparison of the detection probability of different algorithms under different signal-to-noise ratio (SNR) conditions. Simulation experiments verify the performance advantages of the proposed block-based RPCA-STAP algorithm in complex clutter environments, especially in low SNR scenarios. The blue curve represents the RPCA algorithm. Under low SNR conditions (5–10 dB), the RPCA algorithm can hardly detect targets (PD close to 0). Under high SNR conditions (above 20 dB), the detection probability gradually increases, but its performance is still lower than other algorithms. The RPCA algorithm is sensitive to strong clutter leakage and cannot effectively separate target signals under low SNR conditions, resulting in a high false alarm rate. The orange curve represents the DPCA algorithm. The DPCA algorithm begins to show some detection capability in the low to medium SNR range (10–15 dB), but the detection probability increases slowly, eventually approaching 1 at high SNR (above 20 dB). The DPCA algorithm has limited detection capability for low-speed targets and weak clutter suppression capability, therefore it is prone to missed detections under low SNR conditions. The yellow curve represents the STAP algorithm. Under medium SNR conditions (10–15 dB), the STAP algorithm significantly outperforms the DPCA algorithm. As SNR increases, the detection probability rapidly increases, approaching 1 under high SNR conditions. The STAP algorithm effectively suppresses clutter through spatiotemporal two-dimensional filtering, but it has high requirements for the independent and identically distributed (ICD) conditions of the training samples; its performance degrades when the samples are non-uniform. The purple curve represents the ATI-GoDec algorithm. The ATI-GoDec algorithm performs similarly to the STAP algorithm, but converges more slowly under high SNR conditions (above 20 dB), and its detection probability is slightly lower than other algorithms. The GoDec algorithm is sensitive to parameters; improper parameter settings may lead to false alarms. The phase filtering capability of the ATI algorithm has limited effectiveness in detecting low-speed targets. The green curve represents the RPCA-STAP algorithm. The RPCA-STAP algorithm proposed in this invention performs best across the entire SNR range, especially under medium to low SNR conditions (10–15 dB), where its detection probability is significantly higher than other algorithms, and it rapidly converges to 1 under high SNR conditions.

[0106] The RPCA-STAP algorithm combines the clutter separation capabilities of the RPCA algorithm with the strong clutter suppression advantages of the STAP algorithm. Block processing improves the uniformity of the training samples, satisfying the STAP algorithm's requirement for independent and identically distributed (ICD) conditions. The superiority of the RPCA-STAP algorithm presented in this invention is verified, especially under low to medium SNR conditions, significantly improving the accuracy and reliability of target detection, effectively reducing the false alarm rate, and enhancing the detection capability for weak target signals.

[0107] Figure 4 This is a schematic diagram of an experimental scenario provided by an embodiment of the present invention. Figure 5 This is a schematic diagram of a synthetic aperture radar imaging result provided by an embodiment of the present invention. The invention is verified below through X-band real data processing experiments. This embodiment uses a set of measured data from an airborne three-channel X-band SAR radar. Table 1 shows the radar system parameters. The experimental scenario is an urban area, containing eight cooperative targets moving along the radial direction of the radar (e.g., ...). Figure 4 (As shown). Six targets were moving away from the radar, and two targets were moving towards the radar. The experimental area is marked with a red box in the SAR image (e.g., Figure 5 (As shown).

[0108] Table 1

[0109] Parameter Value Bandwidth 40 MHz Carrier frequency 8.85 GHz Pulse repetition frequency 1000 Hz Number of channels 3 Array spacing 0.559m Platform velocity 115 m / s

[0110] Figure 6a This is a schematic diagram of the target detection results of an RPCA algorithm provided in an embodiment of the present invention. The RPCA algorithm detects all cooperative targets (red circles), but generates a large number of false alarms (green circles) in the building area (strong scattering point). Figure 6b This is a schematic diagram of the target detection result of the STAP algorithm provided in an embodiment of the present invention. The STAP algorithm has a low false alarm rate, but due to training sample contamination (such as the target being mixed with clutter samples), there are still some false alarms. Figure 6c This is a schematic diagram of the target detection result of a DPCA algorithm provided in an embodiment of the present invention. The DPCA algorithm shows that the false alarm rate increases significantly in the background of strong clutter. Figure 6d This is a schematic diagram of the target detection result of an ATI-GoDec algorithm provided in an embodiment of the present invention. The ATI-GoDec algorithm eliminates some false alarms through ATI phase filtering, but misses slow targets (such as targets with low radial velocity). Figure 6e This is a schematic diagram of the target detection result of the RPCA-STAP algorithm provided in an embodiment of the present invention. The RPCA-STAP algorithm combines the low-rank decomposition of RPCA with the strong clutter suppression capability of STAP, detects all cooperative targets, and has the lowest false alarm rate.

[0111] Figure 7This is a schematic diagram of the relocation result of a moving target provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the red diamond represents the initial detected target position, and the green circle represents the target position after relocation using multi-channel phase difference information. The final positions of all cooperative targets accurately correspond to the preset movement trajectories, verifying the reliability of the detection results.

[0112] Table 2 lists the running times of different algorithms. According to Table 2, although the RPCA-STAP algorithm adds a STAP module, its running time is only slightly higher than RPCA and significantly lower than the traditional STAP algorithm due to block processing and the ADMM framework.

[0113] Table 2

[0114] Method Run time RPCA 6.68 STAP 30.05 DPCA 0.2 ATI-GoDec 9.26 RPCA-STAP 7.28

[0115] In the above technical solution, the Markov random field image segmentation algorithm is used to classify clutter types in SAR images, ensuring that each sub-image satisfies the independent and identically distributed condition. This effectively solves the problem that traditional techniques are unable to distinguish moving targets from background clutter due to sample contamination or unevenness, which leads to a decline in moving target detection performance. The RPCA-STAP algorithm is constructed, and the strong clutter suppression capability of the STAP algorithm is utilized to effectively reduce the false alarm rate of the RPCA algorithm in strong clutter backgrounds, significantly improving the accuracy and robustness of target detection. The alternating direction multiplier method is used to update the low-rank matrix and sparse matrix obtained from the RPCA-STAP algorithm, which can reduce computational complexity and improve the practicality of the algorithm.

[0116] Figure 8 This is a block diagram of a multi-channel ground moving target detection device based on RPCA-STAP provided in an embodiment of the present invention. The device 800 may include:

[0117] The image segmentation module 801 is used to process the SAR image to be detected according to the Markov random field image segmentation algorithm to obtain the segmented image; wherein, the segmented image includes sub-images of different clutter types; the clutter type of each sub-image is consistent;

[0118] Image processing module 802 is used to process the segmented image according to the pre-built RPCA-STAP algorithm to obtain the low-rank matrix and the sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm;

[0119] The matrix update module 803 is used to update the low-rank matrix and the sparse matrix corresponding to each sub-image by exchanging direction multipliers, so as to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image.

[0120] The target detection module 804 is used to detect moving targets in the SAR image to be detected based on the updated low-rank matrix and the updated sparse matrix of each sub-image, and obtain the position of the moving targets.

[0121] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0122] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0123] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0124] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multi-channel ground moving target detection method based on RPCA-STAP, characterized in that, The method includes: The SAR image to be detected is processed according to the Markov random field image segmentation algorithm to obtain the segmented image; wherein, the segmented image includes sub-images with different clutter types; the clutter type of each sub-image is consistent; The segmented image is processed according to the pre-built RPCA-STAP algorithm to obtain a low-rank matrix and a sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm; The low-rank matrix and sparse matrix corresponding to each sub-image are updated by exchanging direction multipliers, resulting in the updated low-rank matrix and sparse matrix of each sub-image. Based on the updated low-rank matrix and the updated sparse matrix of each sub-image, the moving targets in the SAR image to be detected are detected to obtain the positions of the moving targets; The RPCA-STAP algorithm is expressed as follows: ； in, Represents a low-rank matrix nuclear norm number, Representing a sparse matrix The L1 norm, To control the sparsity of hyperparameters, For adaptive weighting terms, For adaptive weights, A low-rank matrix The generated auxiliary variables, for The conjugate transpose of . Yes The autocorrelation matrix, For adaptive suppression hyperparameters; This represents the constraint condition in the above equation. The constraints representing the Frobenius norm, Indicates the tolerable error size. This represents the segmented image. Represents a low-rank matrix The conjugate transpose of . Used to ensure sparse matrix The modulus length of each column in the model remains consistent. sparse matrix Column number, sparse matrix Total number of columns, The first element used to ensure that the adaptive weights are 1.

2. The multi-channel ground moving target detection method based on RPCA-STAP according to claim 1, characterized in that, The step of updating the low-rank matrix and the sparse matrix corresponding to each sub-image using the exchange-direction multiplier method to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image includes: An augmented Lagrangian function is constructed based on the RPCA-STAP algorithm, the low-rank matrix corresponding to each sub-image, and the sparse matrix corresponding to each sub-image. The augmented Lagrangian function is updated by exchanging direction multipliers to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image.

3. The multi-channel ground moving target detection method based on RPCA-STAP according to claim 2, characterized in that, The augmented Lagrange function is expressed as follows: ； in, Indicates the first penalty factor. Indicates the second penalty factor. Represents the first Lagrange multiplier. Indicates the second Lagrange multiplier. Represents the matrix inner product. Denotes the first Lagrange multiplier term. This represents the second Lagrange multiplier term.

4. The multi-channel ground moving target detection method based on RPCA-STAP according to claim 3, characterized in that, The process of updating the augmented Lagrangian function using the method of exchanging direction multipliers to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image includes: By fixing the variables in the augmented Lagrangian function except for the low-rank matrix, the low-rank matrix is ​​updated by minimizing the augmented Lagrangian function, thus obtaining the updated low-rank matrix for each sub-image. By fixing the variables in the augmented Lagrangian function except for the sparse matrix, the sparse matrix is ​​updated by minimizing the augmented Lagrangian function, thus obtaining the updated sparse matrix for each sub-image.

5. The multi-channel ground moving target detection method based on RPCA-STAP according to claim 4, characterized in that, With all variables in the augmented Lagrangian function fixed except for the low-rank matrix, the process of updating the low-rank matrix by minimizing the augmented Lagrangian function is as follows: 。 6. The multi-channel ground moving target detection method based on RPCA-STAP according to claim 4, characterized in that, The process of updating the sparse matrix by minimizing the augmented Lagrangian function, while keeping all variables in the augmented Lagrangian function fixed, is represented as follows: 。 7. A multi-channel ground moving target detection device based on RPCA-STAP, characterized in that, The device includes: The image segmentation module is used to process the SAR image to be detected according to the Markov random field image segmentation algorithm to obtain the segmented image; wherein, the segmented image includes sub-images with different clutter types; the clutter type of each sub-image is consistent; The image processing module is used to process the segmented image according to the pre-built RPCA-STAP algorithm to obtain a low-rank matrix and a sparse matrix corresponding to each sub-image; wherein, the RPCA-STAP algorithm is obtained based on the RPCA algorithm and the STAP algorithm; The matrix update module is used to update the low-rank matrix and the sparse matrix corresponding to each sub-image by exchanging direction multipliers, so as to obtain the updated low-rank matrix and the updated sparse matrix of each sub-image. The target detection module is used to detect moving targets in the SAR image to be detected based on the updated low-rank matrix and the updated sparse matrix of each sub-image, and to obtain the position of the moving targets; The RPCA-STAP algorithm is expressed as follows: ； in, Represents a low-rank matrix nuclear norm number, Representing a sparse matrix The L1 norm, To control the sparsity of hyperparameters, For adaptive weighting terms, For adaptive weights, A low-rank matrix The generated auxiliary variables, for The conjugate transpose of . Yes The autocorrelation matrix, For adaptive suppression hyperparameters; This represents the constraint condition in the above equation. The constraints representing the Frobenius norm, Indicates the tolerable error size. This represents the segmented image. Represents a low-rank matrix The conjugate transpose of . Used to ensure sparse matrix The modulus length of each column in the model remains consistent. sparse matrix Column number, sparse matrix Total number of columns, The first element used to ensure that the adaptive weights are 1.

Images