Infrared dim target detection method, device, equipment and medium

Abstract

The application belongs to the technical field of target detection, and relates to an infrared dim small target detection method, device, equipment and medium. The method comprises the following steps: acquiring an infrared sequence image and obtaining a space-time tensor model; constructing a three-dimensional structure tensor according to the infrared sequence image to obtain a background edge information tensor; calculating a time-constrained Gaussian curvature filter according to the infrared sequence image to obtain a filter feature map tensor; obtaining a feature weight map according to the background edge information tensor and the filter feature map tensor; constructing a weighted space-time total variation model and defining a non-convex function; constructing an augmented Lagrange function according to the space-time tensor model, the feature weight map, the weighted space-time total variation model and the non-convex function; solving the augmented Lagrange function by using an alternating direction multiplier method until convergence is achieved, to obtain a background tensor and a target tensor, and obtaining an infrared dim small target from the target tensor. The application can accurately detect the infrared dim small target.

Description

Technical Field

[0001] This application relates to the field of target detection technology, and in particular to infrared weak target detection methods, devices, equipment and media. Background Technology

[0002] Infrared target detection holds a unique position in many engineering fields and is widely used. With the development of industrial technology, the high requirements for accurate target detection in navigation and guidance processes are a problem worthy of in-depth research. For long-range imaging, targets are usually small and lack clear shape, texture, and structure. According to existing definitions, in a 256×256 image, small targets typically occupy only 1 to 81 pixels. Furthermore, the radiation intensity of small targets is also weak, which undoubtedly increases the difficulty of detection.

[0003] In existing technologies, based on the number of input samples, mainstream infrared small target detection methods can be divided into two categories: single-frame detection and sequence detection.

[0004] Single-frame detection is the fastest and can fully utilize spatial domain information of the image. Most traditional single-frame detection methods can handle different scenes and have been greatly improved. For example, methods based on time / frequency domain filters are effective in simple backgrounds, especially smooth and continuous background images; methods based on the human visual system can extract targets more easily by distinguishing the intensity differences between the magnified target and the background; and methods based on image component analysis, such as low-rank sparse decomposition methods, are also receiving much attention.

[0005] However, single-frame detection, relying solely on spatial information, is insufficient to handle complex scenes. Multi-frame images, on the other hand, help add motion information, which is crucial for detecting moving targets. However, existing multi-frame image, or sequence detection methods, do not fully utilize motion information, making it difficult to accurately detect small infrared targets. Summary of the Invention

[0006] Therefore, it is necessary to provide a method, apparatus, device, and medium for detecting infrared weak targets, which can accurately detect infrared weak targets, in order to address the above-mentioned technical problems.

[0007] Infrared methods for detecting weak targets include:

[0008] Acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images;

[0009] Based on the infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method; based on the three-dimensional structure tensor, feature values ​​are calculated to obtain a background edge information tensor; based on the infrared sequence images, a time-constrained Gaussian curvature filter is calculated to obtain a filtered feature map tensor; based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained; based on the enhanced local feature tensor map and the target tensor, a feature weight map is obtained.

[0010] Construct a weighted spatiotemporal total variational model and define nonconvex functions;

[0011] Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, an augmented Lagrangian function is constructed.

[0012] The augmented Lagrangian function is solved using the alternating direction multiplier method until convergence, resulting in a background tensor and a target tensor. The infrared weak target is then obtained from the target tensor.

[0013] Infrared weak target detection device, including:

[0014] An acquisition module is used to acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images;

[0015] The processing module is configured to: construct a three-dimensional structure tensor based on the infrared sequence image using the structure tensor method; calculate feature values ​​based on the three-dimensional structure tensor to obtain a background edge information tensor; calculate a time-constrained Gaussian curvature filter based on the infrared sequence image to obtain a filtered feature map tensor; obtain an enhanced local feature tensor map based on the background edge information tensor and the filtered feature map tensor; and obtain a feature weight map based on the enhanced local feature tensor map and the target tensor.

[0016] The modeling module is used to construct a weighted spatiotemporal total variational model and define nonconvex functions;

[0017] The function module is used to construct an augmented Lagrangian function based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function.

[0018] The detection module is used to solve the augmented Lagrangian function using the alternating direction multiplier method until convergence, to obtain the background tensor and the target tensor, and to obtain the infrared weak target from the target tensor.

[0019] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0020] Acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images;

[0021] Based on the infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method; based on the three-dimensional structure tensor, feature values ​​are calculated to obtain a background edge information tensor; based on the infrared sequence images, a time-constrained Gaussian curvature filter is calculated to obtain a filtered feature map tensor; based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained; based on the enhanced local feature tensor map and the target tensor, a feature weight map is obtained.

[0022] Construct a weighted spatiotemporal total variational model and define nonconvex functions;

[0023] Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, an augmented Lagrangian function is constructed.

[0024] The augmented Lagrangian function is solved using the alternating direction multiplier method until convergence, resulting in a background tensor and a target tensor. The infrared weak target is then obtained from the target tensor.

[0025] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0026] Acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images;

[0027] Based on the infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method; based on the three-dimensional structure tensor, feature values ​​are calculated to obtain a background edge information tensor; based on the infrared sequence images, a time-constrained Gaussian curvature filter is calculated to obtain a filtered feature map tensor; based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained; based on the enhanced local feature tensor map and the target tensor, a feature weight map is obtained.

[0028] Construct a weighted spatiotemporal total variational model and define nonconvex functions;

[0029] Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, an augmented Lagrangian function is constructed.

[0030] The augmented Lagrangian function is solved using the alternating direction multiplier method until convergence, resulting in a background tensor and a target tensor. The infrared weak target is then obtained from the target tensor.

[0031] The aforementioned infrared weak target detection method, device, equipment, and medium construct a spatiotemporal tensor model (STT) based on infrared patch images (IPI), which can better reflect the nonlocal correlation of the background and provide more accurate estimation and calculation. Combining a three-dimensional structural tensor and a time-constrained Gaussian curvature filter, an enhanced feature weight map is obtained, which considers temporal and motion information, effectively highlighting moving objects and better estimating the background. A weighted spatiotemporal total variational model (total variational regularization) and a non-convex approximation function are used to optimize the augmented Lagrangian function, and the alternating direction multiplier method is used to solve the optimization problem, effectively suppressing background noise. This application is a target detection method based on low-rank and sparse separation, designing an enhanced local feature spatiotemporal tensor model that fully utilizes three-dimensional temporal features and motion information, and also considers the case of low-intensity targets or non-Gaussian distributed targets. It can achieve accurate detection of infrared weak moving targets, has high reliability, meets engineering requirements, and has wide applicability and high practical value. Attached Figure Description

[0032] Figure 1 This is an application scenario diagram of an infrared weak target detection method in one embodiment;

[0033] Figure 2 This is a flowchart illustrating an infrared weak target detection method in one embodiment;

[0034] Figure 3 This is a schematic diagram of tensor reconstruction in one embodiment;

[0035] Figure 4 This is a flowchart of an infrared weak target detection method in one embodiment;

[0036] Figure 5 This is a comparison chart of the detection results of various methods for scene 1 in one embodiment, where (a) is scene 1, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0037] Figure 6 This is a comparison chart of the detection results of various methods for scene 2 in one embodiment, where (a) is scene 2, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0038] Figure 7 This is a comparison chart of the detection results of various methods for scene 3 in one embodiment, where (a) is scene 3, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0039] Figure 8 This is a comparison chart of the detection results of various methods for scene 4 in one embodiment, where (a) is scene 4, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0040] Figure 9 This is a comparison chart of the detection results of various methods for scene 5 in one embodiment, where (a) is scene 5, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0041] Figure 10 This is a comparison chart of the detection results of various methods for scene 6 in one embodiment, where (a) is scene 6, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0042] Figure 11 This is a comparison chart of the detection results of various methods for scene 7 in one embodiment, where (a) is scene 7, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0043] Figure 12This is a comparison chart of the detection results of various methods for scene 8 in one embodiment, where (a) is scene 8, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0044] Figure 13 This is a comparison chart of the detection results of various methods for scene 9 in one embodiment, where (a) is scene 9, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0045] Figure 14 This is a comparison chart of the detection results of various methods for scene 10 in one embodiment, where (a) is scene 10, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0046] Figure 15 This is a comparison chart of the detection results of various methods for scene 11 in one embodiment, where (a) is scene 11, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0047] Figure 16 This is a comparison chart of the detection results of various methods for scene 12 in one embodiment, where (a) is scene 12, (b) is WSLCM, (c) is MSLSTIPT, (d) is MFSTPT, (e) is FKRW, (f) is NTFRA, (g) is PSTNN, (h) is GST, (i) is ECASTT, (j) is STLDM, (k) is ASTTV, and (l) is SLFSTT;

[0048] Figure 17 This is a structural block diagram of an infrared weak target detection device in one embodiment;

[0049] Figure 18 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0051] The method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 may include, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Server 104 may be a server corresponding to various portal websites or work system backends.

[0052] This application provides a method for detecting weak infrared targets, such as... Figure 2 As shown, in one embodiment, the method is applied to Figure 1 Taking the terminal in the example, the explanation includes:

[0053] Step 202: Obtain infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images.

[0054] Specifically, two or more infrared sequence images (i.e., input images) are acquired, and all infrared sequence images are reconstructed to obtain a sequence image with image patches as local units, i.e., a spacetime tensor model.

[0055] In this step, constructing a spatiotemporal tensor model can highlight the low-rank properties of the background and better adapt to infrared small targets with different distributions, such as... Figure 3 As shown.

[0056] Step 204: Based on the infrared sequence image, construct a three-dimensional structure tensor using the structure tensor method; calculate feature values ​​based on the three-dimensional structure tensor to obtain a background edge information tensor; calculate a time-constrained Gaussian curvature filter based on the infrared sequence image to obtain a filtered feature map tensor; obtain an enhanced local feature tensor map based on the background edge information tensor and the filtered feature map tensor; obtain a feature weight map based on the enhanced local feature tensor map and the target tensor.

[0057] Specifically:

[0058] Based on the aforementioned infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method:

[0059] (1)

[0060] (2)

[0061] (3)

[0062] In the formula, It is a symmetric positive semi-definite matrix. The variance is Gaussian kernel function, For convolution operations, For gradient calculation, The image is a Gaussian-smoothed infrared sequence. For Kronecker product, for Yan Gradient plot of direction, for or direction, for, and For the reason and Two different eigenvalues ​​obtained from directional gradient calculation. , , , These are the different elements in the structure tensor matrix. for and The average eigenvalues, for and The average eigenvalues, and For the reason and Two different eigenvalues ​​obtained from directional gradient calculation. and For the reason and Two different eigenvalues ​​obtained by calculating the directional gradient.

[0063] Based on the 3D structure tensor, eigenvalues ​​are calculated to highlight background edge details, resulting in a background edge information tensor:

[0064] (4)

[0065] (5)

[0066] In the formula, Background edge information map, and Let be two distinct eigenvalues ​​of a symmetric positive semi-definite matrix. To enhance the background edge information tensor, for tensor form, for The minimum value, for The maximum value.

[0067] Based on the infrared sequence images, a time-constrained Gaussian curvature filter is calculated to obtain the filtered feature map tensor:

[0068] (6)

[0069] (7)

[0070] In the formula, This is a Gaussian curvature filtered image. Infrared sequence images, For the minimum projection operator, This is the regularized selective difference plot. This is the filter feature map tensor.

[0071] Based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained:

[0072] (8)

[0073] (9)

[0074] In the formula, For coefficient tensors, for tensor form, To enhance the local feature tensor map.

[0075] Based on the enhanced local feature tensor map and the target tensor, the feature weight map is obtained:

[0076] (10)

[0077] In the formula, For feature weight map, This is a dot product operation. For the target tensor, It is a positive integer.

[0078] In this step, combining the three-dimensional structural tensor and time-constrained Gaussian curvature filtering enables effective feature description.

[0079] Step 206: Construct a weighted spatiotemporal total variational model and define nonconvex functions.

[0080] Specifically:

[0081] Construct a spatiotemporal total variation model and apply adaptive weighting to obtain a weighted spatiotemporal total variation model:

[0082] (11)

[0083] (12)

[0084] (13)

[0085] In the formula, For any tensor, for x Directional difference operator, for y Directional difference operator, for z Directional difference operator, , , t These are the coordinates of different elements in the tensor. For local standard deviation operators, The image is an infrared sequence image after mean filtering. The subscript TV represents the total variational norm, and the subscript 1 represents the L1 norm.

[0086] Define a nonconvex function:

[0087] (14)

[0088] In the formula, It is a very small positive parameter. This is the natural index value.

[0089] In this step, non-convex functions are defined to approximate the 0 norm and low-rank constraints to avoid NP-hard problems.

[0090] Step 208: Construct an augmented Lagrangian function based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the non-convex function.

[0091] Specifically:

[0092] Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, a detection model is constructed:

[0093] (15)

[0094] In the formula, For the core tensor, To assist the tensor, for transpose, For along i Dimensional expansion matrix, For intrinsic tensor sparsity, For pattern i Lower product operation, , , These are the decomposition matrices for modes 1, 2, and 3, respectively. , , For different penalty parameters, and For different auxiliary tensors, i The numbers 1, 2, and 3 represent patterns 1, 2, and 3 respectively, and are used as subscripts. ATV For adaptive total variation norm, subscript F It is the F-norm.

[0095] Based on the detection model, the augmented Lagrangian function is obtained:

[0096] (16)

[0097] In the formula, For zero-norm surrogate functions, Choose 1, 2, 3, For the nuclear norm surrogate function, for according to i The matrix of pattern expansion, middle Pick x, y and z , respectively represent , and , and For two different Lagrange multipliers, and For different penalty parameters.

[0098] In this step, preferably, the feature weight map is reconstructed sequentially, and an augmented Lagrangian function is constructed based on the reconstructed feature weight map.

[0099] Step 210: Solve the augmented Lagrangian function using the alternating direction multiplier method until convergence, to obtain the background tensor and the target tensor, and obtain the infrared weak target from the target tensor.

[0100] Specifically:

[0101] renew :

[0102] (17)

[0103] (18)

[0104] In the formula, For core tensor In the The value in the next iteration For threshold operators, As a reference tensor, for The core tensor in Tucker decomposition , , They are respectively , , transpose, For penalty parameters, For the target tensor in the th case The value in the next iteration To assist the tensor, and For different Lagrange multipliers, As a reference tensor, For Fourier transform, For the transpose of the difference operator, For the number of iterations, It is the conjugate transpose;

[0105] renew :

[0106] (19)

[0107] (20)

[0108] In the formula, For along i Dimensional expansion matrix In the The value in the next iteration To The value obtained by performing singular value decomposition. To Singular value decomposition yields transpose, For reference tensor, for The singular value matrix, For along i Pattern unfolding, In accordance with i Pattern unfolding, For along j 1-dimensional unfolded matrix For along j A 2-dimensional unfolded matrix, where, j 1. j 2 can be 1, 2, or 3 and is not equal to 2. i ,Right now: i When it is 1, j 1. j 2 can only be 2 or 3. i When it is 2, j 1. j 2 can only be 1 or 3. i When it is 3, j 1. j 2 can only be either 1 or 2;

[0109] renew :

[0110] (twenty one)

[0111] (twenty two)

[0112] In the formula, For sparse target tensors In the The value in the next iteration For soft thresholding operators, For core tensor In the The value in the next iteration , , They represent , , In the The value in the next iteration For soft thresholding operators, To set parameters, It is a symbolic function;

[0113] Solve for the Lagrange function, iteratively calculate and update. , as well as until convergence; based on the final update and Obtain the background tensor ( ); based on the background tensor And the final update , obtain the target tensor Based on the target tensor, weak infrared targets are obtained by segmentation.

[0114] In this embodiment, a spatiotemporal tensor model is first constructed, followed by local feature extraction to obtain a feature weight map. A weighted spatiotemporal total variational model is then constructed, and a non-convex function is defined to build a detection model for infrared weak targets. The detection model is then converted into an augmented Lagrangian function and solved. The solution is a background tensor and a target tensor. Finally, the infrared weak target is segmented from the target tensor, as shown below. Figure 4 As shown.

[0115] The aforementioned infrared weak target detection method constructs a spatiotemporal tensor model (STT) based on infrared patch images (IPI), which can better reflect the nonlocal correlation of the background and provide more accurate estimation and calculation. It combines a three-dimensional structure tensor and a time-constrained Gaussian curvature filter to obtain an enhanced feature weight map, which considers temporal and motion information, effectively highlighting moving objects and better estimating the background. A weighted spatiotemporal total variational model (total variational regularization) and a non-convex approximation function are used to optimize the augmented Lagrangian function, and the alternating direction multiplier method is employed to solve the optimization problem, effectively suppressing background noise. This application presents a target detection method based on low-rank and sparse separation, designing an enhanced local feature spatiotemporal tensor model that fully utilizes three-dimensional temporal features and motion information. It also considers the cases of low-intensity targets or non-Gaussian distributed targets, enabling accurate detection of infrared weak moving targets with high reliability, meeting engineering requirements, and possessing wide applicability and high practical value.

[0116] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0117] In one specific embodiment, simulation experiments were conducted using the method of this application (i.e., SLFSTT) and methods in the prior art (i.e., WSLCM, MSLSTIPT, MFSTPT, FKRW, NTFRA, PSTNN, GST, ECASTT, STLDM, ASTTV), and the results were used for, for example... Figures 5 to 16As shown in the figure, the boxes indicate the locations of the true targets, while the circles enclose noise clutter. The figure demonstrates that this method achieves good detection results, exhibits better robustness for detecting small targets in complex scenes, and demonstrates excellent performance.

[0118] This application also provides an infrared weak target detection device, such as Figure 17 As shown, in one embodiment, it includes: an acquisition module 1702, a processing module 1704, a modeling module 1706, a function module 1708, and a detection module 1710, wherein:

[0119] The acquisition module 1702 is used to acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images;

[0120] The processing module 1704 is configured to: construct a three-dimensional structure tensor based on the infrared sequence image using the structure tensor method; calculate feature values ​​based on the three-dimensional structure tensor to obtain a background edge information tensor; calculate a time-constrained Gaussian curvature filter based on the infrared sequence image to obtain a filtered feature map tensor; obtain an enhanced local feature tensor map based on the background edge information tensor and the filtered feature map tensor; and obtain a feature weight map based on the enhanced local feature tensor map and the target tensor.

[0121] Modeling module 1706 is used to construct a weighted spatiotemporal total variational model and define non-convex functions;

[0122] Function module 1708 is used to construct an augmented Lagrangian function based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function.

[0123] The detection module 1710 is used to solve the augmented Lagrangian function using the alternating direction multiplier method until convergence, to obtain the background tensor and the target tensor, and to obtain the infrared weak target from the target tensor.

[0124] Specific limitations regarding the infrared weak target detection device can be found in the limitations of the infrared weak target detection method described above, and will not be repeated here. Each module in the aforementioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the corresponding operations of each module.

[0125] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 18As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for detecting weak infrared targets. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0126] Those skilled in the art will understand that Figure 18 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0128] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0130] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for detecting weak infrared targets, characterized in that, include: Acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images; Based on the infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method; Based on the three-dimensional structure tensor, eigenvalues ​​are calculated to obtain the background edge information tensor; based on the infrared sequence image, a time-constrained Gaussian curvature filter is calculated to obtain the filtered feature map tensor; based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained; based on the enhanced local feature tensor map and the target tensor, a feature weight map is obtained. Construct a weighted spatiotemporal total variational model and define nonconvex functions; Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, an augmented Lagrangian function is constructed. The augmented Lagrangian function is solved using the alternating direction multiplier method until convergence, resulting in a background tensor and a target tensor. The infrared weak target is then obtained from the target tensor. Based on the infrared sequence images, a three-dimensional structure tensor is constructed using the structure tensor method: In the formula, It is a symmetric positive semi-definite matrix. The variance is Gaussian kernel function, For convolution operations, For gradient calculation, The image is a Gaussian-smoothed infrared sequence. For Kronecker product, for Yan Gradient plot of direction, for or direction, and For the reason and Two different eigenvalues ​​obtained from directional gradient calculation. , , , These are the different elements in the structure tensor matrix. for and The average eigenvalues, for and The average eigenvalues, and For the reason and Two different eigenvalues ​​obtained from directional gradient calculation. and For the reason and Two distinct eigenvalues ​​obtained by calculating the directional gradient; Based on the 3D structure tensor, the eigenvalues ​​are calculated to obtain the background edge information tensor: In the formula, Background edge information map, and Let be two distinct eigenvalues ​​of a symmetric positive semi-definite matrix. To enhance the background edge information tensor, for tensor form, for The minimum value, for The maximum value.

2. The infrared weak target detection method according to claim 1, characterized in that, Based on the infrared sequence images, a time-constrained Gaussian curvature filter is calculated to obtain the filtered feature map tensor: In the formula, This is a Gaussian curvature filtered image. Infrared sequence images, For the minimum projection operator, This is the regularized selective difference plot. This is the filter feature map tensor.

3. The infrared weak target detection method according to claim 2, characterized in that, Based on the background edge information tensor and the filtered feature map tensor, an enhanced local feature tensor map is obtained: In the formula, For coefficient tensors, for tensor form, To enhance the local feature tensor map; Based on the enhanced local feature tensor map and the target tensor, the feature weight map is obtained: In the formula, For feature weight map, This is a dot product operation. For the target tensor, It is a positive integer.

4. The infrared weak target detection method according to claim 3, characterized in that, Construct a weighted spatiotemporal total variational model: In the formula, For any tensor, for x Directional difference operator, for y Directional difference operator, for z Directional difference operator, , , t These are the coordinates of different elements in the tensor. For local standard deviation operators, The image is an infrared sequence image after mean filtering. The subscript TV represents the total variational norm, and the subscript 1 represents the L1 norm. Define a nonconvex function: In the formula, It is a very small positive parameter. This is the natural index value.

5. The infrared weak target detection method according to any one of claims 1 to 4, characterized in that, Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, an augmented Lagrangian function is constructed: Based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function, a detection model is constructed: In the formula, For the core tensor, For the target tensor, To assist the tensor, for transpose, For along i Dimensional expansion matrix, For intrinsic tensor sparsity, For pattern i Lower product operation, , , These are the decomposition matrices for modes 1, 2, and 3, respectively. , , For different penalty parameters, and For different auxiliary tensors, i The numbers 1, 2, and 3 represent patterns 1, 2, and 3 respectively, and are used as subscripts. ATV For the adaptive total variational norm, subscript 1 represents the L1 norm, and subscript... F It is the F-norm; Based on the detection model, the augmented Lagrangian function is obtained: In the formula, For zero-norm surrogate functions, Choose 1, 2, 3. For the nuclear norm surrogate function, for according to i The matrix of pattern expansion, middle Pick x, y and z , respectively represent , and , and For two different Lagrange multipliers, This is the penalty parameter.

6. The infrared weak target detection method according to any one of claims 1 to 4, characterized in that, The augmented Lagrangian function is solved using the alternating direction multiplier method until convergence, yielding the background tensor and the target tensor: renew : In the formula, For the core tensor in the first The value in the next iteration For threshold operators, For reference tensor, for The core tensor in Tucker decomposition , , They are respectively , , transpose, and For different penalty parameters, For the target tensor in the th case The value in the next iteration To assist the tensor, and For different Lagrange multipliers, For reference tensor, For Fourier transform, For the transpose of the difference operator, For the number of iterations, It is the conjugate transpose; renew : In the formula, For along i Dimensional expansion matrix In the The value in the next iteration To The value obtained by performing singular value decomposition. To Singular value decomposition yields transpose, For reference tensor, for The singular value matrix, For along i Pattern unfolding, In accordance with i Pattern unfolding, For along j 1-dimensional unfolded matrix For along j A 2-dimensional unfolded matrix, where, j 1. j 2 can be 1, 2, or 3 and is not equal to 2. i ,Right now: i When it is 1, j 1. j 2 can only be 2 or 3. i When it is 2, j 1. j 2 can only be 1 or 3. i When it is 3, j 1. j 2 can only be either 1 or 2; renew : In the formula, For sparse target tensors In the The value in the next iteration For soft thresholding operators, For core tensor In the The value in the next iteration , , They represent , , In the The value in the next iteration For soft thresholding operators, To set parameters, It is a symbolic function; According to the updated and Obtain the background tensor Based on the background tensor and updated Obtain the target tensor .

7. An infrared weak target detection device, characterized in that, The infrared weak target detection method according to any one of claims 1 to 6 includes: An acquisition module is used to acquire infrared sequence images and obtain a spatiotemporal tensor model based on the infrared sequence images; The processing module is configured to: construct a three-dimensional structure tensor based on the infrared sequence image using the structure tensor method; calculate feature values ​​based on the three-dimensional structure tensor to obtain a background edge information tensor; calculate a time-constrained Gaussian curvature filter based on the infrared sequence image to obtain a filtered feature map tensor; obtain an enhanced local feature tensor map based on the background edge information tensor and the filtered feature map tensor; and obtain a feature weight map based on the enhanced local feature tensor map and the target tensor. The modeling module is used to construct a weighted spatiotemporal total variational model and define nonconvex functions; The function module is used to construct an augmented Lagrangian function based on the spatiotemporal tensor model, the feature weight map, the weighted spatiotemporal total variational model, and the nonconvex function. The detection module is used to solve the augmented Lagrangian function using the alternating direction multiplier method until convergence, to obtain the background tensor and the target tensor, and to obtain the infrared weak target from the target tensor.

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

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

Images