A method and device for infrared dim small target detection based on multi-scale background rank

Abstract

The present invention discloses a method and device for infrared dim small target detection based on multi-scale background rank. The method includes: 1) constructing a three-dimensional space-time tensor based on the original thermal infrared image sequence; 2) building an infrared dim small target detection framework based on low-rank sparse decomposition theory; 3) constructing a multi-scale low-rank estimation norm of the background tensor; 4) establishing an infrared dim small target detection model based on multi-scale background rank, and introducing a posterior information reweighting strategy; 5) designing an optimization solution algorithm based on ADMM to solve the model established in step 4) to obtain a target detection result tensor; 6) converting the target detection result tensor into a target detection result sequence to realize infrared dim small target detection. The infrared dim small target detection method based on multi-scale background rank of the present invention can effectively improve the background suppression capability, target detection capability and comprehensive detection performance of infrared dim small targets.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method and device for detecting infrared dim small targets based on multi-scale background rank. Background Art

[0002] Infrared detection and tracking systems have attracted widespread attention in the military and civilian fields due to their strong anti-interference capabilities and the advantages of all-weather and all-day imaging. These systems have been applied to a variety of scenarios, such as security monitoring, night monitoring, aerospace imaging, forest fire prevention, and intelligent transportation systems. However, due to the long distance between the target and the sensor, the target in the infrared image usually exists in a "small" and "weak" form. In addition, atmospheric radiation and atmospheric attenuation will further weaken the visibility of the target, resulting in the lack of clear structure and texture information of the infrared small target. In actual scenarios, highlight interference and pixel-level noise in complex backgrounds make the infrared weak small target detection task more difficult; at the same time, infrared images with low contrast and lack of chromaticity information will hinder the application and transplantation of computer vision methods. In addition, practical applications put forward higher requirements on the real-time performance of infrared small target detection technology.

[0003] In order to meet the above challenges, a large number of infrared dim small target detection methods have been proposed, including technologies based on single-frame detection and multi-frame detection. The single-frame-based method mainly relies on information in the spatial domain, and enhances the saliency of the target by building a mathematical model or designing a filter, while suppressing the background and interference. This type of method has the advantages of high computational efficiency and simple implementation, but its detection effect often drops sharply when dealing with complex backgrounds or strong noise.

[0004] In contrast, multi-frame-based methods make full use of the temporal information of infrared image sequences, and can effectively distinguish targets from background interference by extracting the motion characteristics or temporal change patterns of targets between multiple frames. Such methods perform better in terms of detection accuracy, but often require higher computational costs and are difficult to meet the real-time requirements in practical applications. In addition, existing methods often find it difficult to strike a balance between target detectability (TD) and background suppression (BS), and to further improve TD and BS. Therefore, in order to further improve the practicality of infrared dim small target detection technology, there is an urgent need for an infrared dim small target detection method that can achieve excellent TD and BS performance in complex environments and has fast detection capabilities. Summary of the invention

[0005] In view of the deficiencies in the prior art, the object of the present invention is to provide a method and device for infrared dim small target detection based on multi-scale background rank.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention discloses an infrared dim small target detection method based on multi-scale background rank, which comprises the following steps:

[0008] Step 1): Based on the original thermal infrared image sequence, the image frames are stacked in the time dimension to construct a three-dimensional space-time tensor;

[0009] Step 2): Based on the three-dimensional space-time tensor, build an infrared weak target detection framework based on low-rank sparse decomposition theory;

[0010] Step 3): Decompose the background tensor into multi-scale background sub-tensors; each background sub-tensor is evenly divided in three dimensions according to the number of scales to obtain a block tensor; for each block tensor, calculate the logarithmic sum of the singular values ​​of all its positive slices and define it as the tensor nuclear norm based on log calculation , the tensor nuclear norm of all block tensors of the same background sub-tensor Sum, then perform weighted summation on all background sub-tensors to construct a multi-scale low-rank estimation norm of the background tensor ;

[0011] Step 4): Based on the infrared dim small target detection framework based on low-rank sparse decomposition theory built in step 2), the sparsity reweighting strategy and the posterior information reweighting strategy are introduced to establish an infrared dim small target detection model based on multi-scale background rank;

[0012] Step 5): Design an optimization solution algorithm based on ADMM to solve the model established in step 3) and obtain the target detection result tensor;

[0013] Step 6): Convert the target detection result tensor into a target detection result sequence T as the infrared weak target detection result to realize infrared weak target detection.

[0014] The present invention also discloses an infrared dim small target detection device based on multi-scale background rank for implementing the method, which comprises:

[0015] A space-time tensor construction module is used to construct the infrared image sequence to be detected into a three-dimensional space-time tensor;

[0016] Low-rank sparse decomposition framework building module, used to build an infrared weak target detection framework based on low-rank sparse decomposition theory;

[0017] The multi-scale low-rank estimation norm design module is used to decompose the background tensor into multi-scale background sub-tensors, and then evenly divide each background sub-tensor in three dimensions according to the number of scales to obtain a block tensor. For each block tensor, the logarithm sum of the singular values ​​of all its positive slices is calculated and defined as the tensor nuclear norm based on log calculation. The tensor nuclear norms of all block tensors of the same background sub-tensor are summed, and then a weighted sum is performed on all background sub-tensors to construct a multi-scale low-rank estimation norm of the background tensor to estimate the low-rank characteristics of the background tensor.

[0018] The model building module based on multi-scale background rank is used to build an infrared dim small target detection model based on multi-scale background rank and introduce a posterior information re-weighting strategy;

[0019] Model optimization solution module, used to solve and establish infrared dim small target detection model based on multi-scale background rank;

[0020] The target detection result output module is used to output the target detection result image corresponding to each frame of infrared image in the infrared image sequence.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1) This paper constructs a spatiotemporal tensor to make full use of the spatiotemporal information of the infrared image sequence and carry out the infrared weak small target detection task based on low-rank sparse decomposition. For the background component, a tensor nuclear norm based on multi-scale logarithms is proposed. By performing multi-scale decomposition on the background tensor, the background tensor is represented as the weighted sum of multi-scale sub-tensors. Furthermore, combined with the block tensors of the sub-tensors at each scale, a tensor nuclear norm based on log calculation is designed. , which comprehensively considers the global and local low-rank characteristics of the background tensor at different scales, thereby improving the accuracy of low-rank background estimation.

[0023] 2) Combined with the log-based tensor nuclear norm , estimating the target sparsity Norm, Frobenius norm of estimated noise component, sparsity weighting strategy and posterior information re-weighting strategy, the present invention designs an infrared dim small target detection method based on multi-scale background rank, which provides an effective solution for improving infrared dim small target detection; after the effectiveness verification of the embodiments, the infrared dim small target detection method based on multi-scale background rank established by the present invention can achieve satisfactory target detection capability, background suppression capability and comprehensive detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of constructing a three-dimensional space-time tensor in the present invention;

[0025] Figure 2 It is a structural schematic diagram of an infrared dim small target detection device based on multi-scale background rank in the present invention;

[0026] Figure 3 Schematic diagram of multi-scale decomposition of background tensor;

[0027] Figure 4 This is a sample frame image of a thermal infrared image sequence used for experimental testing;

[0028] Figure 5 The thermal infrared small target detection result obtained by the method proposed by the present invention is an example frame of the thermal infrared image sequence;

[0029] Figure 6 The original image of an example frame of a thermal infrared image and the thermal infrared weak small target detection result diagram after being detected by the method of the present invention, METTR, TSPK, TWTVR, STT-TRNR, SSTTC, 4DTR, FGLR-MCP and AMFUNet. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is described in detail below in conjunction with specific embodiments. Specific embodiments are described below to simplify the present invention. However, it should be recognized that the present invention is not limited to the illustrated embodiments, and various modifications of the present invention are possible without departing from the basic principles, and these equivalent forms also fall within the scope defined by the claims attached to the present application.

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] The idea of ​​the method of the present invention is to construct a three-dimensional spatiotemporal tensor, make full use of the spatiotemporal information of the infrared image sequence to carry out low-rank sparse component decomposition, carry out multi-scale decomposition of the background tensor, obtain the weighted sum of multiple scale background sub-tensors of the original background tensor, and divide the background sub-tensor into several block tensors, thereby designing a tensor nuclear norm based on log calculation that can comprehensively consider the global and local low-rank characteristics of the background tensor at different scales, and realize accurate estimation of low-rank background. In addition, through the tensor nuclear norm based on log calculation, the target sparsity estimation Norm, Frobenius norm of noise component estimation, sparsity weighting strategy and posterior information re-weighting strategy are used to establish an infrared dim small target detection model based on multi-scale background rank, and the alternating direction multiplier method is used to solve the infrared dim small target detection model to realize infrared dim small target detection.

[0033] According to the above ideas, an embodiment of the present invention proposes an infrared dim small target detection method based on multi-scale background rank, which mainly includes the following steps:

[0034] Step 1: Based on the original thermal infrared image sequence, stack the image frames in the time dimension to construct a three-dimensional space-time tensor ;

[0035] Specifically, Figure 1 As shown, based on the original thermal infrared image sequence, it is stacked in the time dimension to form a three-dimensional space-time tensor , the three-dimensional space-time tensor The size is , express The width of express Height, express Thickness;

[0036] Specifically in the embodiment, the constructed three-dimensional space-time tensor The size is ;

[0037] Step 2: Based on the three-dimensional space-time tensor , build an infrared dim small target detection framework based on low-rank sparse decomposition theory;

[0038] Specifically, based on the three-dimensional space-time tensor , the infrared weak small target detection framework based on low-rank sparse decomposition theory is:

[0039] (1)

[0040] in, represents the background tensor, represents the target tensor, represents the noise tensor, represents the rank estimation operator, express norm, represents the square of the Frobenius norm, and Represents the trade-off coefficient.

[0041] Step 3: Decompose the background tensor into multi-scale background sub-tensors; each background sub-tensor is evenly divided in three dimensions according to the number of scales to obtain a block tensor; for each block tensor, calculate the logarithmic sum of the singular values ​​of all its positive slices and define it as the tensor nuclear norm based on log calculation , the tensor nuclear norm of all block tensors of the same background sub-tensor Sum, then perform weighted summation on all background sub-tensors, and construct a multi-scale low-rank estimation norm of the background tensor ;

[0042] Specifically, Figure 3 As shown, the background tensor is decomposed into multiple scales: Decompose into M The weighted sum of the background subtensors of scales, further, m The background subtensor of the scale In each dimension, it is further divided into 2 non-overlapping (m-1) share, , so we get the background subtensor The corresponding 2 3(m-1) Block Tensors , The length values ​​of the first, second, and third dimensions are , the expression of the above multi-scale decomposition is as follows:

[0043] (2)

[0044] in, represents the weight coefficient, represents the background subtensor of the mth scale, express The j-th block tensor of Represents the tensor merge operator;

[0045] Specifically in this embodiment, in order to consider the multi-scale structural information of the background component and reduce the complexity of the algorithm, the background tensor It is divided into the weighted sum of three background sub-tensors, expressed as , , .

[0046] Furthermore, the present invention provides a method for each block tensor , , ,calculate The logarithmic sum of the singular values ​​of all positive slices is defined as the log-based tensor nuclear norm , the calculation formula is:

[0047] (3)

[0048] in, express The pth frontal slice of express The sth singular value of is a normal number.

[0049] Constructing a multi-scale low-rank estimation norm of the background tensor for:

[0050] (4).

[0051] Specifically in the embodiment, M is taken as 3, .

[0052] Step 4: Based on the infrared dim small target detection framework based on low-rank sparse decomposition theory built in step 2), the sparsity reweighting strategy and the posterior information reweighting strategy are introduced to establish an infrared dim small target detection model based on multi-scale background rank;

[0053] Specifically, based on the infrared small target detection framework based on low-rank sparse decomposition theory built in step 2), the sparsity reweighting strategy is introduced, and the sparsity reweighted tensor is calculated as , Represents a positive constant to prevent the denominator from being zero; establish an infrared weak target detection model:

[0054] (5)

[0055] in, represents the Hadamard product;

[0056] Introducing the posterior information reweighting strategy, we obtain the three-dimensional space-time tensor Infrared dim small target detection model based on multi-scale background rank:

[0057] (6)

[0058] in, and represents the weight coefficient, represents the square of the Frobenius norm, express Norm, used for approximation norm, " indicates a symbolic update operator, Indicates The tensor data to be decomposed in the round iteration, Indicates the tensor data to be decomposed in the t+1th iteration. Represents the posterior information in the tth iteration, which is designed as , Represents an identity transformation.

[0059] Step 5: Design an optimization solution algorithm based on ADMM to solve the model established in step 3) and obtain the target detection result tensor ;

[0060] Specifically, the alternating direction method of multipliers (ADMM) algorithm is used to solve the model (6) in step 4), introducing the auxiliary variable , and the equivalent model is

[0061] (7)

[0062] The corresponding augmented Lagrangian function for:

[0063] (8)

[0064] in, is the penalty coefficient, is the Lagrange multiplier;

[0065] Decompose (8) into several sub-problems and update the variables alternately;

[0066] 5.1) The iteration formula is:

[0067] (9)

[0068] in, , and are all unitary matrices, represents a singular value tensor, , represents the inverse fast Fourier transform along the third dimension, , represents the fast Fourier transform along the third dimension; , is a positive constant, if , then , represents the symbolic function, represents the fast Fourier transformed ;if , then ;

[0069] 5.2) The iteration formula is:

[0070] (10)

[0071] 5.3) The iteration formula is:

[0072] (11)

[0073] in, is a soft threshold shrinkage operator, and , is the soft threshold;

[0074] 5.4) The iteration formula is:

[0075] (12)

[0076] 5.5) The iteration formula is:

[0077] (13)

[0078] 5.6) The iteration formula is:

[0079] (14)

[0080] 5.7) The iteration formula is:

[0081] (15)

[0082] (16)

[0083] (17)

[0084] 5.8) The iteration formula is:

[0085] (18)

[0086] in, represents the update coefficient, express The maximum value of

[0087] During the iteration process, the iteration stop condition is: satisfying the relative error , Indicates the threshold, or the value between two consecutive iterations No change;

[0088] Finally, we get the target detection result tensor ;

[0089] Specifically in the embodiment, , , , , , , , , , represents the unit tensor, , , Represents a zero tensor.

[0090] Step 6: Target detection result tensor It is converted into a target detection result sequence T as the infrared weak target detection result to realize infrared weak target detection.

[0091] Corresponding to the aforementioned embodiment of an infrared dim small target detection method based on multi-scale background rank, the present invention also provides an embodiment of an infrared dim small target detection device based on spatiotemporal background reconstruction.

[0092] Figure 2 FIG. 1 is a block diagram of an infrared dim small target detection device based on multi-scale background rank according to an exemplary embodiment. Figure 2 As shown, the device comprises:

[0093] A space-time tensor construction module is used to construct the infrared image sequence to be detected into a three-dimensional space-time tensor;

[0094] Low-rank sparse decomposition framework building module, used to build an infrared weak target detection framework based on low-rank sparse decomposition theory;

[0095] The multi-scale low-rank estimation norm design module is used to decompose the background tensor into multi-scale background sub-tensors, and then evenly divide each background sub-tensor in three dimensions according to the number of scales to obtain a block tensor. For each block tensor, the logarithm sum of the singular values ​​of all its positive slices is calculated and defined as the tensor nuclear norm based on log calculation. The tensor nuclear norms of all block tensors of the same background sub-tensor are summed, and then a weighted sum is performed on all background sub-tensors to construct a multi-scale low-rank estimation norm of the background tensor to estimate the low-rank characteristics of the background tensor.

[0096] The model building module based on multi-scale background rank is used to build an infrared dim small target detection model based on multi-scale background rank and introduce a posterior information re-weighting strategy;

[0097] Model optimization solution module, used to solve and establish infrared dim small target detection model based on multi-scale background rank;

[0098] The target detection result output module is used to output the target detection result image corresponding to each frame of infrared image in the infrared image sequence.

[0099] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0100] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic. The various modules in the device are a kind of logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another unit. Another point is that the connection between the modules shown or discussed can be a communication connection through some interfaces, which can be electrical or other forms. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present application. Ordinary technicians in this field can understand and implement it without paying creative labor. The following takes the disclosed real thermal infrared image sequence as an example to illustrate the specific implementation method to reflect the technical effect of the present invention. The specific steps in the embodiment will not be repeated.

[0101] Example

[0102] The accompanying drawings shown in the embodiments of the present invention can make the purpose, technical solutions and advantages of the present invention more clearly described. It should be noted that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. All equivalent replacements, improvements, etc. made within the method ideas and principles provided by the present invention should be included in the protection scope of the present invention.

[0103] In this embodiment, a public infrared image sequence is used to verify the effective performance of the infrared dim small target detection method based on multi-scale background rank. The data set consists of 75 consecutive infrared image sequences, each with a resolution of 256×256. In this sequence, a drone is flying slowly, forming a clear grayscale contrast with the complex ground background. The background includes ground scenes, telephone poles, complex texture structures, and highlight block areas. There is also strong clutter interference and noise in the image, which further increases the difficulty of target detection. Figure 4 shows an example frame of an infrared image sequence, Figure 5 This is the infrared dim small target detection result obtained by this method. From the detection results, it can be intuitively observed that this method can significantly enhance the target while effectively suppressing background interference and noise components, showing excellent target detection capabilities. In addition, from a quantitative perspective, the performance of the infrared dim small target detection method based on multi-scale background rank is evaluated by the 3D-ROC evaluation index system, as follows:

[0104] 1) Target detection capability:

[0105] 2) Joint target detection capability:

[0106] 3) Background suppression capability:

[0107] 4) Joint background suppression capability:

[0108] 5) Detector effectiveness:

[0109] 6) Joint target detection and background suppression capabilities:

[0110] 7) Target-background signal ratio:

[0111] 8) Comprehensive testing capabilities:

[0112] In order to more objectively verify the effectiveness of the method of the present invention, 8 thermal infrared dim target detection methods are selected for comparison with the method of the present invention. The comparison methods include METTR, TSPK, TWTVR, STT-TRNR, SSTTC, 4DTR, FGLR-MCP and AMFUNet. Table 1 shows the indicators of the 3D-ROC evaluation system for infrared dim target detection results using the comparison methods and the method of the present invention. The bold values ​​and underlined values ​​represent the corresponding optimal performance and suboptimal performance, respectively.

[0113] METTR originated from Cao Z, Kong X, Zhu Q, et al. Infrared dim target detection via mode-k1k2 extension tensor tubal rank under complex ocean environment[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 181: 167-190.

[0114] TSPK is sourced from Pang D, Ma P, Feng Y, et al. Tensor spectral k - support norm minimization for detecting infrared dim and small target against urban backgrounds[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1 - 13.

[0115] TWTVR is sourced from Zhao E, Dong L, Li C, et al. Infrared maritime target detection based on temporal weight and total variation regularization under strong wave interferences[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024.

[0116] STT - TRNR is sourced from Yi H, Yang C, Qie R, et al. Spatial - temporal tensor ring norm regularization for infrared small target detection[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 1 - 5.

[0117] SSTTC is sourced from Xia C, Chen S, Huang R, et al. Separable Spatial - Temporal Patch - Tensor Pair Completion for Infrared Small Target Detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024.

[0118] 4DTR originated from Wu F, Yu H, Liu A, et al. Infrared small target detection using spatiotemporal 4-D tensor train and ring unfolding[J]. IEEETransactions on Geoscience and Remote Sensing, 2023, 61: 1-22.

[0119] FGLR-MCP originated from Liu T, Liu Y, Yang J, et al. Graph Laplacianregularization for fast infrared small target detection[J]. PatternRecognition, 2025, 158: 111077.

[0120] AMFUNet comes from Chung WY, Lee IH, Park C G. Lightweight infrared smalltarget detection network using full-scale skip connection U-Net[J]. IEEEGeoscience and Remote Sensing Letters, 2023, 20: 1-5.

[0121] Figure 6 The original image of the infrared image example frame and the thermal infrared weak target detection result diagram after the method of the present invention, METTR, TSPK, TWTVR, STT-TRNR, SSTTC, 4DTR, FGLR-MCP and AMFUNet detection, where the green area represents the unsuppressed clutter, the red frame represents the correctly detected target area, and the blue frame represents the unsuccessfully detected target area. Figure 6From the visualization results of target detection, it can be seen that TSPK has the problem of missed target detection, TWTVR has the worst background suppression ability, and the "target tail" problem appears in the detection results of SSTTC. The clutter suppression ability of METTR, STT-TRNR, 4DTR and FGLR-MCP is insufficient. AMFUNet can detect the target well in the example frame, but the target shape is distorted; in contrast, the infrared weak target detection method based on multi-scale background rank proposed in the present invention can well suppress the background and noise, and enhance the significance of weak targets. From the quantitative evaluation index results shown in Table 1, the comparison method cannot well achieve the goals of target enhancement and background suppression at the same time, resulting in the comprehensive performance of the detector failing to meet the requirements of practical applications; in contrast, the target detection ability, background suppression ability and comprehensive ability of the method proposed in the present invention can achieve the best effect on the infrared image data set of the embodiment. Combined with the above qualitative analysis and quantitative analysis, the infrared weak target detection method based on multi-scale background rank proposed in the present invention has superior target detection ability, background suppression ability and comprehensive effectiveness.

[0122] Table 1 - Quantitative indicators of the detection results of thermal infrared image example sequences using METTR, TSPK, TWTVR, STT-TRNR, SSTTC, 4DTR, FGLR-MCP, AMFUNet and the method of the present invention

[0123]

[0124] The accompanying drawings shown in the embodiments of the present invention can make the purpose, technical solutions and advantages of the present invention more clearly described. It should be noted that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. All equivalent replacements, improvements, etc. made within the method ideas and principles provided by the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting infrared dim small targets based on multi-scale background rank, characterized in that: The steps include: Step 1): Based on the original thermal infrared image sequence, the image frames are stacked in the time dimension to construct a three-dimensional space-time tensor; Step 2): Based on the three-dimensional space-time tensor, build an infrared weak target detection framework based on low-rank sparse decomposition theory; Step 3): Decompose the background tensor into multi-scale background sub-tensors; each background sub-tensor is evenly divided in three dimensions according to the number of scales to obtain a block tensor; for each block tensor, calculate the logarithmic sum of the singular values ​​of all its positive slices and define it as the tensor nuclear norm based on log calculation , the tensor nuclear norm of all block tensors of the same background sub-tensor Sum, then perform weighted summation on all background sub-tensors to construct a multi-scale low-rank estimation norm of the background tensor ; Step 4): Based on the infrared dim small target detection framework based on low-rank sparse decomposition theory built in step 2), the sparsity reweighting strategy and the posterior information reweighting strategy are introduced to establish an infrared dim small target detection model based on multi-scale background rank; The step 4) is specifically as follows: Based on the infrared small target detection framework based on low-rank sparse decomposition theory built in step 2), the sparsity reweighting strategy is introduced, and the sparsity reweighted tensor is calculated as , Represents a positive constant to prevent the denominator from being zero; establish an infrared weak target detection model: （5） in, represents the Hadamard product; Introducing the posterior information reweighting strategy, we obtain the three-dimensional space-time tensor Infrared dim small target detection model based on multi-scale background rank: （6） in, and represents the weight coefficient, represents the square of the Frobenius norm, express Norm, used to approximate Norm, " " indicates a symbolic update operator, Indicates The tensor data to be decomposed in the round iteration, Represents the tensor data to be decomposed in the t+1th iteration. Represents the posterior information in the tth iteration, which is designed as , represents an identity transformation; Step 5): Design an optimization solution algorithm based on ADMM to solve the model established in step 4) and obtain the target detection result tensor; Step 6): Convert the target detection result tensor into a target detection result sequence T as the infrared weak target detection result to realize infrared weak target detection.

2. The infrared dim small target detection method based on multi-scale background rank according to claim 1 is characterized in that: The step 1) comprises: based on the original thermal infrared image sequence, stacking the image frames in the time dimension direction to construct a three-dimensional space-time tensor , express The width of express Height, express Thickness.

3. The infrared dim small target detection method based on multi-scale background rank according to claim 1 is characterized in that: The step 2) is specifically as follows: Based on three-dimensional space-time tensor , build an infrared small target detection framework based on low-rank sparse decomposition theory: （1） in, represents the background tensor, represents the target tensor, represents the noise tensor, represents the rank estimation operator, express norm, represents the square of the Frobenius norm, and Represents the trade-off coefficient.

4. The infrared dim small target detection method based on multi-scale background rank according to claim 1 is characterized in that: The step 3) is specifically as follows: The background tensor Decomposed into the weighted sum of background sub-tensors of M scales; the background sub-tensor of the mth scale In each dimension, it is further divided into 2 non-overlapping (m-1) share, , get the background subtensor The corresponding 2 3(m-1) Block Tensors , the expression of the above multi-scale decomposition is as follows: （2） in, represents the weight coefficient, represents the background subtensor of the mth scale, express The j-th block tensor of Represents the tensor merge operator; For each block tensor ,calculate The logarithmic sum of the singular values ​​of all positive slices is defined as the log-based tensor nuclear norm , the calculation formula is: （3） in, express The pth frontal slice of express The sth singular value of is a positive constant; Constructing a multi-scale low-rank estimation norm of the background tensor for: （4）。 5. The infrared dim small target detection method based on multi-scale background rank according to claim 4 is characterized in that: The background tensor Decomposed into the weighted sum of background sub-tensors of M scales, including: The background tensor It is divided into the weighted sum of three background sub-tensors, expressed as , , .

6. The infrared dim small target detection method based on multi-scale background rank according to claim 1 is characterized in that: The step 5) is specifically as follows: The ADMM algorithm is used to solve the model (6) in step 4), introducing auxiliary variables , and the equivalent model is （7） The corresponding augmented Lagrangian function for: （8） in, is the penalty coefficient, is the Lagrange multiplier; Decompose (8) into several sub-problems and update the variables alternately , , , , , , , ; 5-1) The iteration formula is: （9） in, , and are all unitary matrices, represents a singular value tensor, , represents the inverse fast Fourier transform along the third dimension, , represents the fast Fourier transform along the third dimension; , is a positive constant, if , then , represents the symbolic function, represents the fast Fourier transformed ;if , then ; 5-2) The iteration formula is: （10） 5-3) The iteration formula is: （11） in, is a soft threshold shrinkage operator, and , is the soft threshold; 5-4) The iteration formula is: （12） 5-5) The iteration formula is: （13） 5-6) The iteration formula is: （14） 5-7) The iteration formula is: （15） （16） （17） 5-8) The iteration formula is: （18） in, represents the update coefficient, express The maximum value of During the iteration process, the iteration stop condition is: satisfying the relative error , Indicates the threshold, or the value between two consecutive iterations No change; At the end of the iteration, the target detection result tensor is obtained .

7. The infrared dim small target detection method based on multi-scale background rank according to claim 1 is characterized in that: The step 6) is specifically as follows: The target tensor obtained at the end of the iteration Reconstructed into infrared target detection result map, the small target detection result map sequence T is obtained to realize thermal infrared small target detection.

8. An infrared dim small target detection device based on multi-scale background rank for implementing the method of claim 1, characterized in that: include: A space-time tensor construction module is used to construct the infrared image sequence to be detected into a three-dimensional space-time tensor; Low-rank sparse decomposition framework building module, used to build an infrared weak target detection framework based on low-rank sparse decomposition theory; The multi-scale low-rank estimation norm design module is used to decompose the background tensor into multi-scale background sub-tensors, and then evenly divide each background sub-tensor in three dimensions according to the number of scales to obtain a block tensor. For each block tensor, the logarithm sum of the singular values ​​of all its positive slices is calculated and defined as the tensor nuclear norm based on log calculation. The tensor nuclear norms of all block tensors of the same background sub-tensor are summed, and then a weighted sum is performed on all background sub-tensors to construct a multi-scale low-rank estimation norm of the background tensor to estimate the low-rank characteristics of the background tensor. The model building module based on multi-scale background rank is used to build an infrared dim small target detection model based on multi-scale background rank and introduce a posterior information re-weighting strategy; Model optimization solution module, used to solve and establish infrared dim small target detection model based on multi-scale background rank; The target detection result output module is used to output the target detection result image corresponding to each frame of infrared image in the infrared image sequence.

Images