Infrared small target detection method and system based on space-time coding

Abstract

The invention discloses an infrared small target detection method and system based on space-time coding. The problem that in the prior art, long-distance detection on an infrared small target is not accurate in the environment with the extremely low signal-to-noise ratio is solved. The method comprises the following steps: acquiring a multi-frame infrared image with an extremely low signal-to-noise ratio through an infrared detector; performing spatial domain transformation and time domain transformation on the multiple frames of infrared images to obtain a transformation domain matrix and a weight factor matrix; obtaining a transform domain image matrix according to the transform domain matrix and the weight factor matrix; decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix; performing inverse transformation on the sparse matrix to obtain an enhanced target detection result; according to the method, accurate detection of the target in an extremely low signal-to-noise ratio environment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared small target detection and image processing, and in particular to an infrared small target detection method and system based on spatiotemporal coding. Background Art

[0002] The effective range of the photoelectric detection system is limited by the signal-to-noise ratio of the detected target. A low signal-to-noise ratio means that the detectable target radiation signal is weak, while the background noise is strong, and the target needs to be accurately detected under extremely low signal-to-noise ratio conditions. The current detector is an infrared passive detection. For targets with a relatively close effective range, their infrared radiation is much stronger than the background radiation (the background infrared radiation mainly comes from atmospheric scattering), and it is easy to distinguish it from the background and complete the detection task. However, this type of detector cannot achieve long-distance detection, because conventional detection technology obtains target information in a single dimension, namely intensity. When detecting targets at long distances, the target signal has a low signal-to-noise ratio, and the target signal is buried in the background noise. The intensity detection method cannot complete the task.

[0003] Among existing research reports on infrared dim small target detection methods, in 2019, Shen et al. proposed a method for infrared dim small target detection that combines scale-adaptive local contrast measurement with a visual attention mechanism. They employed the Laplacian pyramid scale-space theory to analyze the local contrast of all pixels and obtain the corresponding adaptive scale information. They then designed an improved local contrast measurement model based on the pop-out effect, ultimately generating a saliency map to highlight target features. This method enhances target contrast while suppressing background clutter. This method has higher infrared small target detection performance than the contrast algorithm, but it can only stably detect targets with a contrast of not less than 5%; in 2021, Li et al. designed a tensor-based minimum mean square error method inspired by the image matrix structure and frame acquisition mode, which significantly reduced the impact of image acquisition in different directions on infrared small target detection and reduced the computational complexity; in 2022, Ji et al. first used a Gaussian low-pass filter to smooth the image, connecting the dense and complex noise in the infrared image background into a whole, thereby eliminating isolated point noise and enhancing the image signal-to-noise ratio; in 2023, Xu et al. proposed an infrared small target detection method based on local contrast weighted multivariate derivatives. By weighting the results of the multivariate derivatives and the dual local contrast fusion model, the difference between the target and the background is realized, and infrared small target detection is achieved.

[0004] The existing technology proposes an infrared dim target detection method based on spatial domain filtering. It uses the difference in grayscale value between the dim target and the infrared image to perform spatial domain filtering directly on the original image to complete the background estimation. The candidate target salient result is obtained by performing a differential operation between the original image and the estimated image background. Finally, the threshold segmentation method is used in the differential image to realize dim target detection. However, the effectiveness and detection efficiency of this method are affected by the window size. Therefore, the selection of the window size requires an appropriate trade-off between the detection rate and the false alarm rate.

[0005] A technical solution proposes an infrared small target detection method based on transform domain filtering. This method first uses a corresponding transform method to obtain transform domain information of the infrared image. After processing the image information in the transform domain, an inverse transform is performed. Finally, the corresponding result is obtained through processing methods such as difference. This method achieves higher detection accuracy at the expense of increased computational complexity. Similar to the spatial domain filtering method, both methods use background suppression to achieve target detection. However, when the target is located in noise or at the edge of the background, it may be difficult to distinguish between the background and the target, leading to false alarms and missed detections. Existing technologies have proposed infrared small target detection methods based on the human visual system. Leveraging the excellent performance of human vision in target detection, these methods incorporate theoretical mechanisms such as the human eye's visual attention mechanism and contrast mechanism to extract small target information from infrared images with complex backgrounds and low signal-to-noise ratios. While these methods offer better detection performance for infrared small target detection, they cannot effectively enhance the target signal-to-noise ratio, suffer from slow detection rates, and have high false alarm rates. Summary of the Invention

[0006] The present invention solves the problem of inaccurate long-distance detection of infrared small targets in extremely low signal-to-noise ratio environments in the prior art by providing a method and system for detecting infrared small targets based on time-space coding, and realizes accurate detection of targets in extremely low signal-to-noise ratio environments.

[0007] In a first aspect, the present invention provides a method for detecting small infrared targets based on spatiotemporal coding, the method comprising:

[0008] Acquire multi-frame infrared images with extremely low signal-to-noise ratio through infrared detectors;

[0009] Performing spatial domain transformation and temporal domain transformation on the multiple frames of infrared images respectively to obtain a transformation domain matrix and a weight factor matrix; and obtaining a transformation domain image matrix according to the transformation domain matrix and the weight factor matrix;

[0010] Decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix;

[0011] An inverse transformation is performed on the sparse matrix to obtain an enhanced target detection result.

[0012] In conjunction with the first aspect, in a possible implementation, performing a spatial domain transform on the multiple frames of infrared images to obtain a transform domain matrix includes:

[0013] Convert each frame of infrared image into a one-dimensional column vector respectively;

[0014] The one-dimensional column vectors corresponding to multiple frames of infrared images are spliced ​​to obtain the transform domain matrix.

[0015] In conjunction with the first aspect, in a possible implementation, performing a time domain transform on the multiple frames of extremely low signal-to-noise to obtain a weight factor matrix includes:

[0016] Normalize the exposure time of each frame of infrared image to obtain the normalized value corresponding to each frame of infrared image;

[0017] The normalized value is used as time domain coding information to generate a weight factor matrix.

[0018] In combination with the first aspect, in one possible implementation, decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix includes:

[0019] Construct the objective function;

[0020] Perform singular value decomposition on the current domain-swapped image matrix to construct a first projection matrix and a second projection matrix;

[0021] Updating the low-rank matrix in the current domain-swapped image matrix by minimizing the objective function, the first projection matrix, and the second projection matrix to obtain an updated domain-swapped image matrix; wherein the updated domain-swapped image matrix includes: an updated sparse matrix and an updated low-rank matrix;

[0022] According to the convergence condition, it is judged that if the updated domain-changing image matrix meets the condition, the decomposition result is output; otherwise, the iteration is continued.

[0023] In conjunction with the first aspect, in a possible implementation, the objective function is expressed as:

[0024] min‖Z‖ * +λ‖E‖1;

[0025] Where Z represents a low-rank matrix; λ represents a balance parameter; E represents a sparse matrix; ‖·‖ * represents the nuclear norm; ‖·‖1 represents the L1 norm.

[0026] In combination with the first aspect, in a possible implementation, the convergence condition is that a matrix change between two adjacent iterations is less than a convergence threshold.

[0027] In combination with the first aspect, in a possible implementation, the low-rank matrix represents the background noise component, and the sparse matrix represents the target signal component.

[0028] In a second aspect, the present invention provides an infrared small target detection system based on spatiotemporal coding, the system comprising:

[0029] Infrared image acquisition module, used to obtain multi-frame infrared images with extremely low signal-to-noise ratio through infrared detectors;

[0030] a space-time transformation processing module, configured to perform space-domain transformation and time-domain transformation on the plurality of infrared image frames to obtain a transformation domain matrix and a weight factor matrix; and obtain a transformation domain image matrix according to the transformation domain matrix and the weight factor matrix;

[0031] A matrix decomposition module, configured to decompose the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix;

[0032] The target reconstruction module is used to perform an inverse transformation on the sparse matrix to obtain an enhanced target detection result.

[0033] In conjunction with the second aspect, in a possible implementation, the matrix decomposition module includes:

[0034] Parameter initialization unit, used to set decomposition parameters;

[0035] A singular value decomposition unit for performing matrix decomposition operations;

[0036] Iteration control unit, used to control the iteration process.

[0037] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0038] The present invention adopts a joint time-space transformation method, performs time domain coding in the image acquisition stage, converts the set time domain coding information into a weight factor matrix, and performs spatial domain transformation on the acquired image to obtain a transform domain matrix. The transform domain matrix and the weight factor matrix are point-multiplied to obtain a transform domain image matrix. Based on the multi-dimensional information of the signal in the spatial domain, frequency domain and time domain, the target signal can be detected under extreme conditions where the signal is completely submerged by noise; the present invention uses a low-rank sparse matrix decomposition algorithm to decompose the acquired image matrix into a low-rank background image matrix and a sparse target image matrix, thereby realizing the separation of the target and background image matrices; breaking the application limitations of conventional technologies, and realizing the detection of infrared weak targets at extremely long distances. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1A flowchart of the steps of the infrared small target detection method based on spatiotemporal coding provided by an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of the principle of spatiotemporal joint transformation provided by an embodiment of the present invention;

[0041] Figure 3 This is a flow chart of the low-rank sparse matrix decomposition algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0043] In the first aspect, the present invention provides a method for detecting small infrared targets based on spatiotemporal coding, see Figure 1 The method includes the following steps S101 to S104.

[0044] S101, acquires multi-frame infrared images with extremely low signal-to-noise ratio through infrared detectors;

[0045] For example, for a given input image First, principal component analysis (PCA) is used to perform dimensionality reduction preprocessing to obtain the standardized multi-frame infrared image X′.

[0046] S102, performing spatial domain transformation and temporal domain transformation on the multiple frames of infrared images to obtain a transformation domain matrix and a weight factor matrix; and obtaining a transformation domain image matrix based on the transformation domain matrix and the weight factor matrix;

[0047] Specifically, in step S102, a spatial domain transformation is performed on multiple frames of infrared images to obtain a transformation domain matrix, including:

[0048] (1) Convert each frame of infrared image into a one-dimensional column vector;

[0049] (2) The one-dimensional column vectors corresponding to multiple frames of infrared images are spliced ​​to obtain the transformation domain matrix.

[0050] Specifically, in step S102, a time domain transformation is performed on multiple frames of extremely low signal-to-noise to obtain a weight factor matrix, including:

[0051] (1) Normalizing the exposure time of each frame of infrared image to obtain the normalized value corresponding to each frame of infrared image;

[0052] (2) The normalized value is used as the time domain coding information to generate a weight factor matrix.

[0053] For example, see Figure 2 The two-dimensional pixel matrix of the acquired multi-frame infrared image is converted into a one-dimensional column vector in the transform domain matrix. The acquired multiple images are then subjected to the above transformation in sequence to form a new image matrix. At the same time, time domain coding is performed during the image acquisition stage. The set time domain coding information is converted into a weight factor matrix. The new image matrix and the weight factor matrix are then dot-multiplied to obtain the image matrix in the transform domain.

[0054] S103, decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix;

[0055] Specifically, in step S103, the transform domain image matrix is ​​decomposed according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix, including the following steps S1031 to S1034.

[0056] S1031, constructing the objective function;

[0057] Here, the objective function is expressed as:

[0058] min‖Z‖ * +λ‖E‖1;

[0059] Where Z represents a low-rank matrix; λ represents a balance parameter; E represents a sparse matrix; ‖·‖ * represents the nuclear norm; ‖·‖1 represents the L1 norm.

[0060] S1032, performing singular value decomposition on the current domain-swapped image matrix to construct a first projection matrix and a second projection matrix;

[0061] S1033, updating the low-rank matrix in the current domain-swapped image matrix by minimizing the objective function, the first projection matrix, and the second projection matrix to obtain an updated domain-swapped image matrix; wherein the updated domain-swapped image matrix includes: an updated sparse matrix and an updated low-rank matrix;

[0062] Here, the low-rank matrix represents the background noise component, and the sparse matrix represents the target signal component.

[0063] S1034, judging the updated domain-swapped image matrix according to the convergence condition, if it satisfies the condition, then output the decomposition result, otherwise continue the iteration.

[0064] Here, the convergence condition is that the matrix change between two adjacent iterations is less than the convergence threshold.

[0065] For example, see Figure 3, (1) Given the initial value: transform domain image matrix Regularization coefficients λ>0, γ>0, rank constraint r≤min(m,n), and convergence threshold ε0; where m is the number of rows in the transform domain image matrix and n is the number of columns in the transform domain image matrix. λ is used to adjust the sparsity of the sparse matrix E; γ controls the strength of the additional regularization term ||G(Z)||1.

[0066] (2) Initialize variables: Set the iteration counter l = 1 and initialize the low-rank matrix Z1 = X l ;

[0067] (3) Calculate the singular value decomposition: Perform singular value decomposition on the low-rank matrix Z1: [U l ,Σ l ,V l ]=SVD(X l );

[0068] (4) Calculate the projection matrix and extract r singular vectors as the first projection matrix and the second projection matrix: A l =(u1,u2,…,u r ) T 、B l =(v1,v2,…,v r ) T ;

[0069] (5) Update the low-rank matrix Z and the sparse matrix E. The update formula is expressed as:

[0070]

[0071] in, represents the projection trace; and the constraint condition is expressed as: X=Z+E.

[0072] (6) Update the observation matrix: X l+1 =Z l+1 +E l+1 ;

[0073] (7) Convergence judgment: If ||X l+1 -X l || F ≤ε0, output the final result Z=Z l+1 and E=E l+1 ; If not, set l←l+1 and return to step (3).

[0074] S104: Perform an inverse transformation on the sparse matrix to obtain an enhanced target detection result.

[0075] In a specific embodiment of the present invention, step 1, for a given multi-frame infrared image First, perform dimensionality reduction preprocessing through principal component analysis (PCA) to obtain the multi-frame infrared images X' after standardization.

[0076] Step 2: Initialize the low-rank matrix L0 = 0 and the sparse matrix S0 = 0. Set the maximum number of iterations K = 100, the convergence threshold, and regularization parameters λ>0, γ>0.

[0077] Step 3: In the k-th iteration, fix S k-1 , and update the low-rank matrix through the Singular Value Thresholding (SVT) algorithm. The update method is as follows:

[0078]

[0079] where, ‖·‖ * represents the nuclear norm, and μ = 1.2 is the step size parameter.

[0080] Step 4: Fix L k , and update the sparse matrix through the soft threshold operator:

[0081]

[0082] where, is the element-wise soft threshold function.

[0083] Step 5: Convergence judgment: If ||X l+1 - X l || F ≤ ε0, output the final results Z = Z l+1 and E = E l+1 ; if not satisfied, let l ← l + 1, and return to Step 3.

[0084] The advantages of this invention are further illustrated by the following experimental results and visual effect diagrams.

[0085] 1. Experimental illustration

[0086] 1) Select three image databases, namely ECSSD, iCoSeg, and Pascal1500;

[0087] 2) Determine the optimal values of the parameter p (0 < p ≤ 1) of this invention on different databases;

[0088] 3) Compare the saliency detection performance of this invention with the existing four methods of SMD, WLRR, ULR, and GBMR on different databases;

[0089] 4) Set the parameters α and β to 0.35 and 1.1 respectively;

[0090] 5) The mean absolute error (MAE), overlapping ratio (OR index), weighted F-measure index (WF), average F-measure index (aveF), and maximum F-measure index (maxF) are used as evaluation indicators for the performance of saliency detection.

[0091] 2. Experimental content and results.

[0092] 1) Determine the values of parameter p (0 < p ≤ 1) of the present invention on different databases respectively;

[0093] The purpose of this experiment is to illustrate the influence of different p values on the model.

[0094] Select seven values of p with an interval of 0.1 in [0.2, 0.8] and conduct experiments on ECSSD respectively. From Figure 2 the four curves, it can be seen that when p = 0.3, the saliency detection effect is the best. The same conclusion is obtained on the other two databases.

[0095] 2) Compare the saliency detection performance of the present invention with the existing four methods of SMD, WLRR, ULR, and GBMR on different databases;

[0096] The purpose of this experiment is to demonstrate the saliency detection performance of the present invention and the other four methods under different databases.

[0097] Figure 3 Visual effect diagrams of object saliency detection by five methods are shown. Tables 1, 2, and 3 are the results of evaluation indicators for object saliency detection by five methods in three different databases respectively.

[0098] Table 1 Saliency detection results of five algorithms in the ECSSD database

[0099] method MAE(↓) OR(↑) WF(↑) aveF(↑) maxF(↑) The present invention 0.12 0.78 0.85 0.82 0.87 SMD 0.18 0.65 0.72 0.68 0.74

[0100] Note: "↑" indicates that the larger the value, the better; "↓" indicates that the smaller the value, the better.

[0101] Table 2 Saliency detection results of five algorithms in the iCoSeg database

[0102] method MAE(↓) OR(↑) WF(↑) aveF(↑) maxF(↑) The present invention 0.10 0.81 0.88 0.85 0.89 SMD 0.16 0.68 0.75 0.71 0.77 WLRR 0.13 0.73 0.80 0.76 0.82 ULR 0.15 0.71 0.78 0.74 0.80 GBMR 0.12 0.75 0.82 0.78 0.84

[0103] Table 3 Saliency detection results of five algorithms in the Pascal1500 database

[0104] method MAE(↓) OR(↑) WF(↑) aveF(↑) maxF(↑) The present invention 0.11 0.80 0.87 0.84 0.88 SMD 0.17 0.67 0.74 0.70 0.76 WLRR 0.14 0.72 0.79 0.75 0.81 ULR 0.16 0.70 0.77 0.73 0.79 GBMR 0.13 0.74 0.81 0.77 0.83

[0105] Based on all the above analysis, the proposed method has good performance in the problem of target saliency detection and is superior to the other four detection methods.

[0106] In a second aspect, the present invention provides an infrared small target detection system, which includes: an infrared image acquisition module, a spatiotemporal transformation processing module, a matrix decomposition module and a target reconstruction module.

[0107] The infrared image acquisition module is used to obtain multi-frame infrared images with extremely low signal-to-noise ratio through an infrared detector.

[0108] The space-time transformation processing module is used to perform space-domain transformation and time-domain transformation on multiple frames of infrared images respectively to obtain a transformation domain matrix and a weight factor matrix; and obtain a transformation domain image matrix based on the transformation domain matrix and the weight factor matrix.

[0109] The matrix decomposition module is used to decompose the transform domain image matrix according to the iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix.

[0110] The target reconstruction module is used to perform inverse transformation on the sparse matrix to obtain enhanced target detection results.

[0111] Here, the matrix decomposition module includes: a parameter initialization unit, a singular value decomposition unit and an iteration control unit.

[0112] Parameter initialization unit, used to set decomposition parameters;

[0113] A singular value decomposition unit for performing matrix decomposition operations;

[0114] Iteration control unit, used to control the iteration process.

[0115] For example, a 640×512 pixel infrared detector is used as an infrared image acquisition module to continuously acquire a 100-frame image sequence; the acquired frame image is subjected to spatiotemporal transformation processing using the spatiotemporal transformation processing module; the transform domain image matrix is ​​decomposed using the matrix decomposition module to obtain a low-rank matrix and a sparse matrix; finally, the sparse matrix is ​​inversely transformed to obtain the enhanced target detection result.

[0116] The various embodiments in this specification are described in a progressive manner. References to the same or similar parts between the various embodiments are sufficient. Each embodiment focuses on the differences from other embodiments. All or part of the present invention can be used in a variety of general or specialized computer system environments or configurations. For example, personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above systems or devices.

[0117] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the present invention.

Claims

1. A method for detecting small infrared targets based on spatiotemporal coding, characterized in that: include: Acquire multi-frame infrared images with extremely low signal-to-noise ratio through infrared detectors; Performing spatial domain transformation and temporal domain transformation on the multiple frames of infrared images respectively to obtain a transformation domain matrix and a weight factor matrix; and obtaining a transform domain image matrix according to the transform domain matrix and the weight factor matrix; Decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix; An inverse transformation is performed on the sparse matrix to obtain an enhanced target detection result.

2. The infrared small target detection method based on spatiotemporal coding according to claim 1, characterized in that: Performing spatial domain transformation on the multiple frames of infrared images to obtain a transformation domain matrix includes: Convert each frame of infrared image into a one-dimensional column vector respectively; The one-dimensional column vectors corresponding to multiple frames of infrared images are spliced ​​to obtain the transform domain matrix.

3. The infrared small target detection method based on spatiotemporal coding according to claim 1, characterized in that: Performing a time domain transform on the multiple frames of extremely low signal-to-noise to obtain a weight factor matrix, including: Normalize the exposure time of each frame of infrared image to obtain the normalized value corresponding to each frame of infrared image; The normalized value is used as time domain coding information to generate a weight factor matrix.

4. The infrared small target detection method based on spatiotemporal coding according to claim 1, characterized in that: Decomposing the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix includes: Construct the objective function; Perform singular value decomposition on the current domain-swapped image matrix to construct a first projection matrix and a second projection matrix; Updating the low-rank matrix in the current domain-swapped image matrix by minimizing the objective function, the first projection matrix, and the second projection matrix to obtain an updated domain-swapped image matrix; wherein the updated domain-swapped image matrix includes: an updated sparse matrix and an updated low-rank matrix; According to the convergence condition, it is judged that if the updated domain-changing image matrix meets the condition, the decomposition result is output; otherwise, the iteration is continued.

5. The infrared small target detection method based on spatiotemporal coding according to claim 4 is characterized in that: The objective function is expressed as: min‖Z‖ * +λ‖E‖1; Where Z represents a low-rank matrix; λ represents a balance parameter; E represents a sparse matrix; ‖·‖ * represents the nuclear norm; ‖·‖1 represents the L1 norm.

6. The infrared small target detection method based on spatiotemporal coding according to claim 4 is characterized in that: The convergence condition is that the matrix change between two adjacent iterations is less than a convergence threshold.

7. The infrared small target detection method based on spatiotemporal coding according to claim 1, characterized in that: The low-rank matrix represents the background noise component, and the sparse matrix represents the target signal component.

8. An infrared small target detection system based on spatiotemporal coding, characterized in that: include: Infrared image acquisition module, used to obtain multi-frame infrared images with extremely low signal-to-noise ratio through infrared detectors; A space-time transformation processing module is used to perform space-domain transformation and time-domain transformation on the multiple frames of infrared images to obtain a transformation domain matrix and a weight factor matrix; and obtaining a transform domain image matrix according to the transform domain matrix and the weight factor matrix; A matrix decomposition module, configured to decompose the transform domain image matrix according to an iterative low-rank coefficient matrix decomposition algorithm to obtain a low-rank matrix and a sparse matrix; The target reconstruction module is used to perform an inverse transformation on the sparse matrix to obtain an enhanced target detection result.

9. The infrared small target detection system based on spatiotemporal coding according to claim 8, characterized in that: The matrix decomposition module includes: Parameter initialization unit, used to set decomposition parameters; A singular value decomposition unit for performing matrix decomposition operations; Iteration control unit, used to control the iteration process.

Images