Space-based external radiation source radar moving target detection method and device based on tensor decomposition

Abstract

The invention discloses a tensor decomposition-based space-based external radiation source radar moving target detection method and device. The method comprises the following steps: acquiring a multi-frame continuous moving target RD original image sequence processed by a space-based external radiation source radar system; filtering the original image sequence based on a pre-constructed annular filtering template to obtain a background estimation image sequence; performing normalization processing on the original image sequence by using the background estimation image sequence to obtain a final image sequence; performing stacking processing on the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; constructing a target detection model of a fourth-order space-time tensor based on the low-rank performance of the background, the sparsity of the target, the noise constraint and the target trajectory smoothness constraint; solving the target detection model to obtain a fourth-order target tensor; and reconstructing the fourth-order target tensor into a multi-frame continuous target image sequence to obtain a target detection result. According to the invention, the detection capability of the moving target can be greatly improved, and the false alarm rate and the missed alarm rate of the target are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, and in particular relates to a space-based passive ESR radar moving target detection method and device based on tensor decomposition. BACKGROUND

[0002] Passive radar, also known as passive radar, is a dual / multi-base radar system that detects targets by receiving and processing the echo signals of existing electromagnetic waves reflected by targets. Traditional passive radar mainly relies on ground-based radiators such as broadcast stations and communication stations. However, due to the geographical distribution of such radiators, the flexibility of system deployment is often insufficient. In recent years, with the rapid development of satellite technology, the number of space-based radiators has increased significantly. Benefiting from its wide coverage and rich data sources, passive radar technology based on space-based radiators has become a research hotspot. Common space-based radiators mainly include global navigation satellite signals, synthetic aperture radar (SAR) satellite signals, digital satellite television signals, and other low / middle / high-orbit radiator satellite signals. Space-based passive radiators have the following advantages: first, they can work all day long and are less affected by external environment; second, due to the high orbit height, the illumination range is wide, and the flexibility of receiver deployment is greatly improved; in addition, space-based radiators are highly secure and have stronger battlefield survivability.

[0003] The space-based passive ESR radar system uses the transmitted signals of the radiator satellites for target detection. However, the ultra-long range of space-based passive ESR results in a generally low signal power on the ground, which is difficult to meet the application requirements of subsequent target detection and tracking. Although single-frame long-time coherent accumulation technology can improve the target signal-to-noise ratio (SNR) to some extent, the gain of such algorithm is limited and the computational complexity is significantly increased when the radar system is observed for a long time (such as a staring observation scenario based on high-orbit radiators). However, in such scenarios, it is easy to obtain continuous multi-frame Range-Doppler (RD) image data. Therefore, how to efficiently utilize high-frame-rate multi-frame data to realize robust detection of weak moving targets has become a key problem in current research.

[0004] Therefore, there is an urgent need for a space-based passive ESR radar moving target detection method and device based on tensor decomposition to solve the above problems. SUMMARY

[0005] The present application provides a space-based passive ESR radar moving target detection method and device based on tensor decomposition, which can greatly improve the detection capability of moving targets and reduce the false alarm rate and the missed alarm rate of targets. The technical solution is as follows:

[0006] On the one hand, a space-based passive ESR radar moving target detection method based on tensor decomposition is provided, the method comprising:

[0007] obtain a plurality of continuous moving target RD original image sequences processed by a space-based passive coherent location radar system;

[0008] filter the original image sequences based on a pre-constructed ring filter template to obtain a background estimation image sequence;

[0009] normalize the original image sequences using the background estimation image sequence to obtain a final image sequence;

[0010] stack the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor and a noise tensor;

[0011] construct a target detection model of the fourth-order space-time tensor based on low-rank of the background, sparsity of the target, noise constraint and target trajectory smooth constraint;

[0012] solve the target detection model to obtain a fourth-order target tensor;

[0013] reconstruct the fourth-order target tensor into a plurality of continuous target image sequences to obtain a target detection result.

[0014] In another aspect, a space-based passive coherent location radar moving target detection device based on tensor decomposition is provided, and the device comprises:

[0015] an acquisition unit configured to acquire a plurality of continuous moving target RD original image sequences processed by a space-based passive coherent location radar system;

[0016] a filtering unit configured to filter the original image sequences based on a pre-constructed ring filter template to obtain a background estimation image sequence;

[0017] a normalization unit configured to normalize the original image sequences using the background estimation image sequence to obtain a final image sequence;

[0018] a stacking unit configured to stack the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor and a noise tensor;

[0019] a construction unit configured to construct a target detection model of the fourth-order space-time tensor based on low-rank of the background, sparsity of the target, noise constraint and target trajectory smooth constraint;

[0020] a solving unit configured to solve the target detection model to obtain a fourth-order target tensor;

[0021] The reconstruction unit is used to reconstruct the fourth-order target tensor into a multi-frame continuous target image sequence to obtain a target detection result.

[0022] On the other hand, a computer device is provided, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the above-mentioned method for detecting moving targets by space-based external radiation source radar based on tensor decomposition.

[0023] On the other hand, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned method for detecting moving targets by space-based external radiation source radar based on tensor decomposition are implemented.

[0024] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for detecting moving targets by space-based exo-radiation source radar based on tensor decomposition.

[0025] The embodiment of the present invention provides a method for detecting moving targets of space-based external radiation source radar based on tensor decomposition. First, by introducing a ring filter template to perform pixel normalization preprocessing on the RD image sequence, the background energy of each pixel can be effectively estimated and the interference of uneven brightness on the detection standard can be suppressed, the signal-to-noise ratio scale of different regions is unified, and the robustness and adaptability of subsequent moving target detection are improved. Secondly, when constructing the target detection model of the fourth-order space-time tensor, the movement of the target is fully considered, and the target trajectory smoothing constraint is introduced. The space-time characteristics such as time domain information, local information and global information of the radar image sequence can be fully utilized, and the target enhancement capability and background suppression capability are better, thereby improving the target detection capability and reducing the false alarm rate and missed alarm rate of the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 This is a flow chart of a method for detecting moving targets by space-based exo-radiation radar based on tensor decomposition, provided by one embodiment of the present invention;

[0028] Figure 2 This is a structural diagram of a space-based exo-radiation source radar moving target detection device based on tensor decomposition provided by one embodiment of the present invention;

[0029] Figure 3 is a hardware architecture diagram of a computer device provided by an embodiment of the present application;

[0030] Figure 4 is a fourth-order space-time tensor construction process schematic diagram provided by an embodiment of the present application;

[0031] Figure 5 is a two-dimensional schematic diagram of the 8th frame of original images provided by an embodiment of the present application;

[0032] Figure 6 is a three-dimensional schematic diagram of the 8th frame of original images provided by an embodiment of the present application;

[0033] Figure 7 is a target image of the 8th frame of original images recovered using an existing 4D-TR tensor decomposition method;

[0034] Figure 8 is a target image of the 8th frame of original images recovered using the method of the present application. DETAILED DESCRIPTION

[0035] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings of the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0036] The specific implementation of the method of the present application will be described in detail below.

[0037] Referring to Figure 1 , the present application provides a space-based external radiation source radar moving target detection method based on tensor decomposition, which comprises the following steps:

[0038] Step 100: obtaining a plurality of frames of continuous moving target RD original image sequences processed by a space-based external radiation source radar system;

[0039] Step 102: performing filtering processing on the original image sequences based on a pre-constructed ring filter template to obtain a background estimation image sequence;

[0040] Step 104: performing normalization processing on the original image sequences by using the background estimation image sequence to obtain a final image sequence;

[0041] Step 106: performing stacking processing on the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor and a noise tensor;

[0042] In step 108, a target detection model of a four-order space-time tensor is constructed based on the low rank of the background, the sparsity of the target, the noise constraint, and the target trajectory smoothness constraint.

[0043] In step 110, the target detection model is solved to obtain a four-order target tensor.

[0044] In step 112, the four-order target tensor is reconstructed into a sequence of multiple frames of target images to obtain a target detection result.

[0045] In this embodiment, first, pixel normalization preprocessing is performed on the RD image sequence by introducing a ring-shaped filtering template, which can effectively estimate the background energy of each pixel and suppress the interference of uneven brightness on the detection standard, unify the signal-to-noise ratio scale of different regions, and improve the robustness and adaptability of subsequent moving target detection. Secondly, in constructing the target detection model of the four-order space-time tensor, the motion of the target is fully considered, and a target trajectory smoothness constraint is introduced, which can fully utilize the space-time characteristics of the time domain information, local information, and global information of the radar image sequence, has better target enhancement capability and background suppression capability, thereby improving the target detection capability and reducing the false alarm rate and the missing alarm rate of the target.

[0046] The execution manner of each step shown in the above embodiment is described below. Figure 1

[0047] First, for step 100:

[0048] The number of frames of the RD original image sequence can be determined as needed, such as 10 frames, which is not limited in the present application.

[0049] For step 102:

[0050] The ring-shaped filtering template includes a kernel region and a background region arranged outside the kernel region; the kernel region is located at the center of the ring-shaped filtering template, and the size of the kernel region covers the target to be detected.

[0051] In some embodiments, the ring-shaped filtering template is defined as:

[0052]

[0053] In the formula, Mk(i,j) represents the pixel value at (i,j) after normalization of the ring-shaped filtering template; and σ(i,j) represents the pixel value at (i,j) before normalization of the ring-shaped filtering template.

[0054] In this embodiment, by using the above ring-shaped filtering template, only the peripheral pixels are retained for estimating the background energy, thereby avoiding the interference of the moving target energy with the accuracy of the background estimation. It is assumed that the size of the outer frame of the ring-shaped filtering template is W​r ×W c , respectively represent the template in the Doppler dimension (row) and the distance dimension (column) window size, the kernel area is K in length and width respectively r and K c The background area is the remaining pixels except the kernel area.

[0055] Regarding step 104:

[0056] Normalization of the original image sequence using the background estimation image sequence is performed based on the following formula:

[0057]

[0058] Where M i represents the final image after normalization of the i-th image; M i-before represents the i-th original image; B i-est represents the i-th background estimation image; i = 1, 2…T, where T is the total number of frames in the image sequence.

[0059] In this step, by normalizing the pixels in the image, the detection threshold offset problem caused by inconsistent background brightness can be effectively suppressed, and a unified detection benchmark can be formed in different areas, which is conducive to the subsequent enhancement and detection of moving targets based on tensor sparse decomposition.

[0060] Regarding step 106, it includes:

[0061] Use a sliding window of preset size to traverse each frame image and obtain N block images corresponding to each frame image;

[0062] The first frame image is used as the reference image. For each block image in the reference image, the block images at the same spatial position in the subsequent multiple frames are stacked in the time domain to form a local three-dimensional tensor. This process is repeated until each block image is traversed to obtain N local three-dimensional tensors.

[0063] Stack N local three-dimensional tensors in the fourth dimension to obtain the final fourth-order space-time tensor.

[0064] In this step, if Figure 4 Figure 1 shows a schematic diagram of the fourth-order space-time tensor construction process. First, assume there are T frames of RD images and the size of the sliding window is W×W. Then, use this sliding window to traverse each frame from left to right and from top to bottom. The step size of each sliding window in rows and columns is W. The single frame image is divided into N block images by the sliding window.

[0065] After obtaining N block images, the first frame image is used as a reference, and the block images at the same spatial position in the subsequent T-1 frame images are stacked into a local three-dimensional tensor according to the time domain. Its dimension is W x W x T, repeating the above steps for all the remaining block images in the first frame will result in N local 3D tensors

[0066] Finally, stack the N local 3D tensors in the fourth dimension (the fourth dimension is the direction from the first block image to the last block image) to obtain the final fourth-order space-time tensor The dimension is W x W x T x N, which is the final fourth-order space-time tensor.

[0067] For step 108, it includes:

[0068] Based on the low-rank of the background, the sparsity of the target and the noise constraint, a basic detection model of the fourth-order space-time tensor is constructed;

[0069] The rank of the background tensor in the basic detection model is approximated by using the tensor ring decomposition kernel norm, and a target trajectory smoothing constraint is introduced into the basic detection model to obtain the final target detection model.

[0070] In this step, it can be considered that the fourth-order space-time tensor composed of the original RD image is composed of three parts of background tensor, target tensor and noise tensor, and accordingly a moving target detection model based on fourth-order tensor decomposition is established. The tensor ring decomposition norm is selected to convexly approximate the rank of the tensor, and the Frobenius norm is used for energy constraint on the noise term, so as to obtain the basic detection model of the fourth-order space-time tensor.

[0071] In some embodiments, the basic detection model is as follows:

[0072]

[0073] In the formula, denotes taking the minimum value on the variable ; rank(·) denotes the rank of the tensor; denotes the fourth-order space-time tensor.

[0074] When solving the above basic detection model, the rank of the background tensor can be approximated by using the tensor ring (TR) decomposition kernel norm, that is, The subscript is 2, which is half of the order of the tensor used (fourth order), and also indicates that the tensor can be changed into two matrices by TR decomposition. At the same time, in order to facilitate subsequent solving, the auxiliary variable

[0075] Finally, the target trajectory smoothing regular term is introduced to obtain the final target detection model.

[0076] In some embodiments, the target detection model is:

[0077]

[0078] wherein, denotes the minimum value of the variable and U n , respectively, and U n denote the background tensor, the target tensor, the noise tensor and the auxiliary variable, respectively; λ1, λ2 and λ3 are the regularization coefficients of the target term, the noise term and the target trajectory constraint term, respectively; ||·||1 denotes the L1 norm; denotes the square of the F norm; ||·||F * denotes the kernel norm; denotes the fourth-order space-time tensor; wherein, the subscript 2 denotes taking one half of the order of the fourth-order tensor used, and also indicates that the fourth-order tensor can be changed into two matrices through TR decomposition; δ n denotes the weight factor (i.e., the tensor TR decomposition coefficient), and the sum of each weight factor is 1, and δ1= δ2= 0.5 is preferably taken; SD(·) denotes the difference summation operator; and denote the target tensor at t and t+1 time, respectively.

[0079] In this step, denotes the target tensor the L1 norm of the difference summation along the time frame dimension, and by introducing this term, the target motion can be constrained, and the smoothness of the target trajectory can be maintained.

[0080] For step 110, the alternating direction multiplier (ADMM) method is preferably adopted to solve the target detection model, so as to obtain the low-rank fourth-order background tensor and the sparse fourth-order target tensor The specific solving process is as follows:

[0081] (1) In order to decouple the L1 norm constraint term and the motion constraint regularization term, the auxiliary variable is introduced in the above target detection model, and then the objective function can be rewritten as:

[0082]

[0083] Next, the above objective function is rewritten into the form of an augmented Lagrangian function:

[0084]

[0085] wherein, denote the Lagrange multipliers corresponding to the three equality constraint terms, respectively; ρ1, ρ (2,n), ρ3 respectively represent the penalty coefficient of the corresponding item.

[0086] (2) According to the calculation principle of the alternating direction multiplier method, the above augmented Lagrange function is decomposed into the following 5 sub-problems. When solving a variable, other variables are regarded as known values, and the solution is obtained by alternating iteration until the convergence condition is met. When the k+1 iteration is performed, the update formula of each sub-problem can be expressed as follows, the right side of the equation takes the known value of the variable at the k time, and is no longer specifically marked out:

[0087] U n Sub-problem solution:

[0088]

[0089] Sub-problem solution:

[0090]

[0091] Sub-problem solution:

[0092]

[0093] Sub-problem solution:

[0094]

[0095] Sub-problem solution:

[0096]

[0097] The update of the Lagrange multiplier is as follows:

[0098]

[0099] In the above formula, the superscripts k and k+1 respectively represent the parameter values corresponding to the kth and k+1th iterations.

[0100] When the above iteration meets the convergence condition, the iteration is stopped, and the required fourth-order background tensor and the fourth-order target tensor It should be noted that the convergence condition is preferably: Of course, the user can also independently set the convergence condition according to the needs, and the present application is not limited thereto.

[0101] It should be further noted that because the main purpose of the algorithm is to decompose the target tensor, only the specific form of the analytical solution of the sub-problem is derived and given above, and the forms of the analytical solutions of the remaining four sub-problems are not described again, but are only represented by the argmin mathematical symbol.

[0102] Finally, for step 102, by reconstructing the fourth-order target tensor, an image sequence containing only the target can be obtained, thereby completing the target detection task.

[0103] Furthermore, by solving the target detection model, we can also obtain a fourth-order background tensor. Similarly, by reconstructing the fourth-order background tensor, we can obtain an image sequence containing only low-rank background, thus completing the background detection task.

[0104] In order to verify the effectiveness of the method of the present application, the inventors verified the method of the present application with the following examples.

[0105] The system simulation and algorithm parameter settings corresponding to this embodiment are shown in Table 1.

[0106] Table 1 System simulation and algorithm parameter settings

[0107]

[0108] In addition, the outer frame size W of the ring filter template r ×W c is 19×19, and the kernel area size is K r ×K c Since the input image dimension is 200×200 and the sliding window W is 10×10, the single frame image is divided into 400 block images by the sliding window. Correspondingly, the fourth-order space-time tensor The dimensions are 10×10×10×400.

[0109] like Figure 5 and Figure 6 As shown in Figure 2, the 2D and 3D schematic diagrams of the original image of frame 8 are shown. As can be seen from the figure, the original image is mainly composed of clutter background at zero Doppler frequency and moving point targets at 100Hz Doppler frequency. Therefore, the main task of target tensor decomposition is to separate the target from the background.

[0110] Figure 7 This is the target image of the eighth frame of the original image, restored using the existing 4D-TR tensor decomposition method. As can be seen from the figure, a large number of clutter false alarms remain in the target image. This is because the existing 4D-TR tensor decomposition method does not fully consider the continuity of the target's motion trajectory between frames, and mistakenly decomposes some interference from the clutter background into the target area.

[0111] Figure 8 This is the processing result obtained by using the fourth-order space-time tensor decomposition method based on motion constraints proposed by the present invention. As can be seen from the figure, the false alarm of the restored target image is greatly reduced. In addition, Figure 8The target position is marked by a red box, which is approximately at the 100th distance sampling point and the Doppler frequency of 100 Hz, which is consistent with the actual situation of the input data, and verifies the effectiveness of the method.

[0112] As shown in Figure 2 , Figure 3 , the embodiment of the application provides a space-based E-SINRADAR moving target detection device based on tensor decomposition. The device embodiment can be realized by software, or realized by hardware or a combination of software and hardware. From the hardware layer, as shown in Figure 2 , a hardware architecture diagram of a computing device where the space-based E-SINRADAR moving target detection device based on tensor decomposition provided by the embodiment of the application is located, in addition to the processor, memory, network interface, and non-volatile memory shown in Figure 2 , the computing device where the device in the embodiment is usually also composed of other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, as shown in Figure 3 , as a logically meaningful device, it is formed by the CPU of the computing device where it is located reading the corresponding computer program in the non-volatile memory into the memory for running.

[0113] Please refer to Figure 3 , the embodiment of the application provides a space-based E-SINRADAR moving target detection device based on tensor decomposition, which comprises:

[0114] The acquisition unit 300 is used for acquiring a plurality of frames of continuous moving target RD original image sequences processed by a space-based E-SINRADAR system;

[0115] The filtering unit 302 is used for filtering the original image sequence based on a pre-constructed ring-shaped filtering template to obtain a background estimation image sequence;

[0116] The normalization unit 304 is used for normalizing the original image sequence by using the background estimation image sequence to obtain a final image sequence;

[0117] The stacking unit 306 is used for stacking the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor, and a noise tensor;

[0118] The construction unit 308 is used for constructing a target detection model of the fourth-order space-time tensor based on the low-rank property of the background, the sparsity of the target, the noise constraint, and the target trajectory smoothness constraint;

[0119] The solving unit 310 is used for solving the target detection model to obtain a fourth-order target tensor;

[0120] The reconstruction unit 312 is configured to reconstruct the fourth-order target tensor into a plurality of frames of continuous target image sequences to obtain a target detection result.

[0121] In some embodiments, the annular filtering template comprises a kernel region and a background region arranged outside the kernel region; the kernel region is located at the center of the annular filtering template, and the size of the kernel region covers the target to be detected.

[0122] In some embodiments, the annular filtering template is defined as:

[0123]

[0124] In the formula, Mk(i,j) represents the pixel value of the annular filtering template after normalization at (i,j); and σ(i,j) represents the pixel value of the annular filtering template before normalization at (i,j).

[0125] In some embodiments, the normalization unit 304 is based on the following formula:

[0126]

[0127] In the formula, M i represents the i-th normalized final image; M i-before represents the i-th original image; B i-est represents the i-th background estimation image; i = 1, 2…T, and T is the total number of frames of the image sequence.

[0128] In some embodiments, the stacking unit 306 is configured to perform the following operations:

[0129] Using a sliding window of a preset size to traverse each frame of image to obtain N block images corresponding to each frame of image;

[0130] Taking the first frame of image as a reference image, for each block image in the reference image, the block images at the same spatial position in the subsequent multiple frames of image are stacked in the time domain as a local three-dimensional tensor; and the same is repeated until each block image is traversed, thereby obtaining N local three-dimensional tensors;

[0131] Stacking the N local three-dimensional tensors in the fourth dimension to obtain a final fourth-order space-time tensor.

[0132] In some embodiments, the construction unit 308 is configured to perform the following operations:

[0133] Based on the low-rank property of the background, the sparsity of the target, and the noise constraint, a basic detection model of the fourth-order space-time tensor is constructed;

[0134] The nuclear norm of tensor ring decomposition is used to approximate the rank of the background tensor in the basic detection model, and the target trajectory smoothness constraint is introduced into the basic detection model to obtain the final target detection model.

[0135] In some embodiments, the target detection model is:

[0136]

[0137] Where, Indicates in variables and U n Take the minimum value, and U n denote the background tensor, target tensor, noise tensor and auxiliary variable respectively; λ1, λ2 and λ3 are the regularization coefficients of target term, noise term and target trajectory constraint term respectively; ‖·‖1 denotes the L1 norm; represents the square of the F norm; || || * represents the nuclear norm; represents the fourth-order space-time tensor; In the equation, the subscript 2 indicates that the order of the fourth-order tensor used is half, which also means that the fourth-order tensor can be transformed into two matrices through TR decomposition; δ n represents the weight factor; SD(·) represents the difference sum operator; and Represent the target tensors at time t and t+1 respectively.

[0138] It should be noted that the tensor decomposition-based space-based exo-radiation radar moving target detection device provided in the above embodiment is merely an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the tensor decomposition-based space-based exo-radiation radar moving target detection device provided in the above embodiment and the tensor decomposition-based space-based exo-radiation radar moving target detection method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0139] The embodiment of the present application also provides a computer device, please refer to Figure 3 The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the space-based exo-radiation source radar moving target detection method based on tensor decomposition provided by the above-mentioned method embodiments.

[0140] The embodiment of the present application further provides a computer readable storage medium, and at least one instruction, at least one program, a code set or an instruction set are stored on the computer readable storage medium, the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to realize the space-based passive bistatic radar moving target detection method based on tensor decomposition provided by each method embodiment.

[0141] The embodiment of the present application further provides a computer program product, and the computer program product comprises a computer program, the computer program is read by a processor of a computer device from a computer readable storage medium, and the processor executes the computer program, so that the computer device executes the space-based passive bistatic radar moving target detection method based on tensor decomposition described in any one of the above embodiments.

[0142] For the convenience of description, the above system or device is described as various modules or units in terms of functions respectively. Of course, the functions of each unit can be realized in one or more software and / or hardware in the implementation of the present application.

[0143] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware platforms. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments of the present application.

[0144] Finally, it should be noted that, in this document, relational terms such as first and second, and the like can only be used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations exist in any actual relationship or order. Moreover, the terms "comprise", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a list of elements does not exclude other elements not explicitly listed, or other elements inherent to such process, method, article or device. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0145] The above merely describes the preferred embodiments of the present application, and it should be pointed out that, for those skilled in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for detecting moving targets of space-based exo-radiation radar based on tensor decomposition, characterized in that: The method comprises: Acquire a multi-frame continuous moving target RD original image sequence processed by a space-based external radiation source radar system; Performing filtering on the original image sequence based on a pre-constructed ring filter template to obtain a background estimation image sequence; Normalizing the original image sequence using the background estimation image sequence to obtain a final image sequence; The final image sequence is stacked based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor, and a noise tensor; Based on the low rank of the background, the sparsity of the target, the noise constraint and the smoothness constraint of the target trajectory, the target detection model of the fourth-order space-time tensor is constructed; Solving the target detection model to obtain a fourth-order target tensor; The fourth-order target tensor is reconstructed into a multi-frame continuous target image sequence to obtain a target detection result.

2. The method according to claim 1, characterized in that The annular filter template includes a kernel area and a background area arranged outside the kernel area; the kernel area is located at the center of the annular filter template, and the size of the kernel area covers the target to be detected.

3. The method according to claim 2, characterized in that The ring filter template is defined as: Where Mk(i,j) represents the normalized pixel value of the ring filter template at (i,j); σ(i, j) represents the pixel value of the ring filter template at (i, j) before normalization.

4. The method according to claim 1, wherein The normalization processing of the original image sequence by using the background estimation image sequence is performed based on the following formula: Where M i represents the final image after normalization of the i-th image; M i-before represents the i-th original image; B i-est represents the i-th background estimation image; i = 1, 2…T, where T is the total number of frames in the image sequence.

5. The method according to claim 1, wherein The stacking process of the final image sequence based on the fourth-order space-time tensor method to obtain a fourth-order space-time tensor includes: Use a sliding window of preset size to traverse each frame image and obtain N block images corresponding to each frame image; The first frame image is used as the reference image. For each block image in the reference image, the block images at the same spatial position in the subsequent multiple frames are stacked in the time domain to form a local three-dimensional tensor. This process is repeated until each block image is traversed to obtain N local three-dimensional tensors. Stack N local three-dimensional tensors in the fourth dimension to obtain the final fourth-order space-time tensor.

6. The method according to claim 1, characterized in that The target detection model of the fourth-order space-time tensor is constructed based on the low rank of the background, the sparsity of the target, the noise constraint, and the target trajectory smoothness constraint, including: Based on the low rank of the background, the sparsity of the target and the noise constraint, a basic detection model of the fourth-order space-time tensor is constructed; The rank of the background tensor in the basic detection model is approximated by the nuclear norm of the tensor ring decomposition, and the target trajectory smoothness constraint is introduced into the basic detection model to obtain the final target detection model.

7. The method according to claim 6, characterized in that The target detection model is: Where, Indicates in variables and U n Take the minimum value, and U n denote the background tensor, target tensor, noise tensor and auxiliary variable respectively; λ1, λ2 and λ3 are the regularization coefficients of target term, noise term and target trajectory constraint term respectively; ‖·‖1 denotes the L1 norm; represents the square of the F norm; || || * represents the nuclear norm; represents the fourth-order space-time tensor; In the example, the subscript 2 indicates that the order of the fourth-order tensor used is half, indicating that the tensor Through TR decomposition, it becomes two matrices; δ n represents the weight factor; SD(·) represents the difference sum operator; and Represent the target tensors at time t and t+1 respectively.

8. A space-based external radiation source radar moving target detection device based on tensor decomposition, characterized in that: The device comprises: An acquisition unit is used to acquire a multi-frame continuous moving target RD original image sequence processed by a space-based external radiation source radar system; A filtering unit, configured to filter the original image sequence based on a pre-constructed ring filter template to obtain a background estimation image sequence; a normalization unit, configured to perform normalization processing on the original image sequence using the background estimation image sequence to obtain a final image sequence; a stacking unit, configured to perform stacking processing on the final image sequence based on a fourth-order space-time tensor method to obtain a fourth-order space-time tensor; the fourth-order space-time tensor is composed of a background tensor, a target tensor, and a noise tensor; A construction unit, configured to construct a target detection model of the fourth-order space-time tensor based on the low rank of the background, the sparsity of the target, the noise constraint, and the target trajectory smoothness constraint; A solving unit, configured to solve the target detection model to obtain a fourth-order target tensor; The reconstruction unit is used to reconstruct the fourth-order target tensor into a multi-frame continuous target image sequence to obtain a target detection result.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-7.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images