CN109934815B - Tensor recovery infrared small target detection method combined with ATV constraint - Google Patents
Tensor recovery infrared small target detection method combined with ATV constraint Download PDFInfo
- Publication number
- CN109934815B CN109934815B CN201910202771.1A CN201910202771A CN109934815B CN 109934815 B CN109934815 B CN 109934815B CN 201910202771 A CN201910202771 A CN 201910202771A CN 109934815 B CN109934815 B CN 109934815B
- Authority
- CN
- China
- Prior art keywords
- tensor
- target
- image
- background
- objective function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a tensor recovery infrared small and weak target detection method combined with ATV constraint, and relates to the field of infrared image processing and target detection; the method comprises the following steps of 1: constructing a third-order tensor of the original image; step 2: constructing a prior information weight tensor of an original image; and step 3: using the tensor logDet function and the tensor l 1 Norm, combining with ATV constraint, constructing an objective function, inputting a third-order tensor and a prior information weight tensor into the objective function, and solving the objective function by using ADMM to obtain a background tensor and an objective tensor; and 4, step 4: reconstructing a background image and a target image according to the background tensor and the target tensor; and 5: segmenting the target image to output a target detection result; the method solves the problems of high false alarm rate in infrared weak and small target detection and local optimality caused by nuclear norm due to the fact that the existing method is easily influenced by background edges and noise, improves the target detection and background suppression capability, and improves the accuracy of target detection.
Description
Technical Field
The invention relates to the field of infrared image processing and target detection, in particular to a tensor recovery infrared small and weak target detection method combined with ATV constraint.
Background
The infrared imaging technology has the characteristics of non-contact property, strong capability of capturing details and the like, and realizes the detection of continuous long-distance targets day and night without being influenced by obstacles such as smoke, fog and the like; the Infrared search and tracking IRST (Infrared search and track) system is widely applied to the fields of military, civil use and the like, wherein the Infrared weak and small target detection technology is used as a basic function of the IRST system and has important significance in Infrared search, infrared early warning and long-distance target detection. However, due to the lack of texture and structural information of the target in the infrared band, the influence of long distance, complex background and various clutter, the infrared target is often in a spot or point shape and even submerged in the background, which makes the detection of the infrared weak and small target extremely difficult.
Infrared small and weak target detection techniques fall into two broad categories: the technology for detecting the weak and small targets based on a single frame and the technology for detecting the weak and small targets based on multiple frames have the advantages that due to the fact that the technology for detecting the weak and small targets based on the multiple frames needs to be combined with the multiple frames to capture the motion tracks of the targets and eliminate noise interference, extremely large calculation amount and storage amount are needed, requirements on hardware are high, and application in practical engineering is few. Currently, the commonly used detection methods based on a single frame are classified into the following three categories:
(1) Background suppression: the background suppression method is based on the assumption of background consistency in the infrared image, a filter is adopted to predict the background of the infrared image, then the background is subtracted from the original image, and finally threshold segmentation is carried out to detect the dim target. Maximum median filtering, maximum mean filtering, top-hat transformation, two-dimensional least mean square filtering, etc. all belong to the category of background suppression. Although this type of method is simple to implement, the background suppression method is very susceptible to noise clutter due to the assumption that noise does not conform to consistency, resulting in poor suppression of most infrared images with low signal-to-noise ratio.
(2) Visual saliency: the Human Visual System HVS (Human Visual System) involves three mechanisms, contrast, visual attention and eye movement, of which the most involved are the contrast mechanism, i.e. assuming that in the infrared image, the object is the most prominent object. For example, a gaussian difference filter calculates a saliency map by using two different gaussian filters, and detects and identifies a target; the method based on local contrast utilizes the characteristics that the local contrast of a small neighborhood containing a target is high, but the local contrast of a background area of the target which is not contained is low, and achieves the aim of detection by computing a local contrast map, highlighting the target and restraining the background. When the infrared image conforms to the assumption of visual saliency, the method can obtain excellent effect, but in an actual application scene, the assumption is difficult to meet, for example, when a salient false alarm source exists, the false detection problem is difficult to overcome, and the accuracy rate is low.
(3) Separating a target background: the method utilizes the non-local autocorrelation of the infrared image background and the sparsity of the target to convert the target detection problem into an optimization problem; the method can be further divided into a method based on an ultra-complete dictionary and low-rank representation and a method based on low-rank background and sparse target restoration. The first method needs to construct an ultra-complete dictionary with different target sizes and shapes in advance by a Gaussian intensity model, the process of constructing the target dictionary is complicated, the detection result is greatly influenced by the dictionary, and the Gaussian intensity model is not applicable any more if the target sizes and shapes are changed greatly; in the second method, a low-rank original block Image can be obtained by means of an IPI (acquired Path-Image) model, and then a background and a target Image are recovered by optimizing a target function by means of the characteristic of target sparsity, so that a detection result is finally obtained; the second method is excellent in effect, but has the following two problems: 1. strong edges, partial noise and false alarm sources are also sparse, so that the detection accuracy is reduced; 2. since the process of objective function optimization requires iteration, it is difficult to achieve real-time performance.
In the current information explosion era, the dimensionality of data is not limited to one dimension and two dimensions, the processing difficulty is increasing day by day, and tensors are used for expressing multidimensional information; in practice, tensors are a general concept of multidimensional arrays, such as one-dimensional arrays, commonly referred to as vectors, and two-dimensional arrays, commonly referred to as matrices. Robust Principal Component Analysis (RPCA) overcomes the defect that the Robust principal Component Analysis is easily influenced by abnormal points, is more Robust, and is widely applied to the fields of image completion, image denoising, face recognition and the like at present; however, RPCA can only be used directly for processing two-dimensional matrices, and if high-dimensional data is to be processed, the high-dimensional data needs to be converted into two-dimensional data, and then converted into a high-dimensional space after the processing is completed. This process is not only cumbersome, but also completely destroys the inherent structure of the data and is inefficient. To be able to process high dimensional data more flexibly, tensor-based techniques are gradually developed, where Tensor Recovery (Tensor Recovery) can make use of more data information (structure, color, time, etc.) and performs better than RPCA on sparse low rank decomposition. Tensor robust principal component analysis (TRPC A) is a key technology in Tensor recovery technology, is high-order expansion of RPCA, and is proposed by Goldfarb and Qin. Given a known tensorAnd is known to +>Can be decomposed into:
wherein the content of the first and second substances,is a low rank tensor,. Epsilon.is a sparse tensor, based on >>Solving for>The problem of e and e is a tensor recovery problem.
A Total Variation (TV) model is a well-known partial differential equation denoising model, and since an image detail part and noise have great similarity, it is difficult to protect the detail part while denoising the image. Osher et al proposed the concept of total variation in 1992, and the model can effectively protect the image edge while denoising. TV has proven to preserve important edges and corners of the image, often as a regularization term when accurate estimation of image discontinuities is required. In other words, TV represents the smoothness of a given image, and it is also widely used for image decomposition, which can decompose an image into two parts: one part is the uncorrelated random pattern and the other part is the sharp edge and piecewise smooth components. By minimizing the TV of the image, the smooth inner surface of the image will be preserved while maintaining sharp edges. The TV model includes Isotropic Total Variation (ITV) and Anisotropic Total Variation (ATV), but the ATV is more and more applied to the fields of image denoising, image reconstruction, etc. because the edge-preserving capability of the ATV is better than that of the ITV. Given a graph X, without loss of generality, the ATV is defined as follows:
wherein the content of the first and second substances,and &>Representing the horizontal and vertical two-dimensional finite difference operators, respectively.
In order to improve the detection capability of the Infrared small and weak target, in consideration of that the traditional Infrared small and weak target detection Method only considers local characteristics of an image and the optimization Method only considers non-local autocorrelation characteristics of the image, the prior document proposes an RIPT (weighted independent pitch-sensor Model) Model, that is, on the basis of a block Tensor Model, an objective function is constructed by combining local and non-local characteristics of the Infrared image, and an Alternating Direction Multiplier Method (ADMM) is used for solving the objective function. In most cases, RIPT has better background suppression and target enhancement capabilities, but the tensor kernel norm adopted by RIPT is the kernel norm and SNN (Sum of Nuclear Norms), and the document "a new conditional relaxation for tensor completion" indicates that SNN is not the optimal convex approximation of tensor rank, all singular values in the kernel norm are given the same weight, and in an actual scene, the singular values of target content and noise are different, so RIPT causes a local optimal solution, and increases the false alarm rate in a target image. Also, local structure weights in RIPTThe edge of the object is highlighted at the same time as the background edge, so that the object shape of the detection result is reduced, and even the object cannot be detected. Therefore, there is a need for an infrared small target detection method that combines tensor recovery and ATV to overcome the above problems.
Disclosure of Invention
The invention aims to: the invention provides a tensor recovery infrared small target detection method combined with ATV constraint, which overcomes the problems of high false alarm rate in infrared small target detection and local optimality caused by nuclear norm, which are easily influenced by background edges and noise, of the existing method, improves the target detection and background inhibition capability, and improves the accuracy of target detection.
The technical scheme adopted by the invention is as follows:
a tensor recovery infrared small target detection method combined with ATV constraint comprises the following steps:
step 1: constructing a third-order tensor of the original image;
step 2: extracting prior information of an original image, and constructing a prior information weight tensor;
and step 3: using tensor logDet function and tensor l 1 Norm, combining with ATV constraint, constructing an objective function, inputting a third-order tensor and a prior information weight tensor into the objective function, and solving the objective function by using ADMM to obtain a background tensor and an objective tensor;
and 4, step 4: reconstructing a background image and a target image according to the background tensor and the target tensor;
and 5: and performing self-adaptive threshold segmentation on the target image to determine the position of the target and outputting a target detection result.
Preferably, the step 1 comprises the steps of:
step 1.1: obtaining an original imageWherein m and n represent the length and width of the image, respectively;
step 1.2: traversing an original image D by adopting a sliding window w with the size of p multiplied by p and according to the step length of s;
step 1.3: taking the small image block in the sliding window w as a front slice every time, and establishing a third-order tensor after sliding for q times
Preferably, the step 2 comprises the steps of:
Wherein, K ρ A Gaussian kernel function representing the variance ρ representing a convolution operation, D σ Means that the variance of the original image is σ: (>0) The gaussian smoothing filter of (1) is used,represents a kronecker product, -is present>Indicating a gradient is taken over>Is shown by D σ The gradient in the x-direction is such that, device for combining or screening>Represents D σ Gradient in the y-direction, J 11 Replacement>J 12 Substitute K ρ *I x I y ,J 21 Substitute K ρ *I x I y ,J 22 Replacement->
Wherein, an indicates a Hadamard product;
W m =max(λ 1 ,λ 2 );
W p =W cs *W m ;
To W p Normalization was done as follows:
wherein w min And w max Respectively represent W p Minimum and maximum values of (d);
step 2.6: according to a normalized prior information matrix W p Constructing a priori information weight tensorThe construction method comprises the following steps: traversing W with a sliding window W of size p p Taking the small image block in the sliding window w as a front section every time, and forming a third-order tensor which is the prior information weight tensor->
Preferably, the constructing the objective function in step 3 includes the following steps:
step a1: third order tensorComprising a low rank tensor pick>And sparse tensor pick>For separating low rank tensor pick>And sparse tensor pick>Taking tensor logDet function as regularization term of low-rank tensor, tensor l 1 The norm is used as a regularization term of the sparse tensor, and an objective function is constructed by combining ATV constraint, wherein the formula is as follows:
where η > 0 represents a small regularization constant, λ and β represent equilibrium coefficients, logDet (-) represents the logDet function of the tensor, and hasMeans for>Represents->I (i is more than or equal to 1 and less than or equal to p) th singular value->Is/>The ith frontal slice of the tensor obtained by discrete fourier transform along the third dimension, where l below denotes the same meaning), i | · | | survival 1 The representation tensor l 1 Norm (i.e., sum of singular values of all elements in the tensor), | ·| luminance HTV Represents an anisotropic total variation constraint, which is defined as->Wherein-> And &>Two-dimensional finite difference operators which respectively represent the horizontal and the vertical;
step a2: order toRepresenting a sparse weight tensor, defining a weight tensor based on the sparse weight and the prior information weight tensor>The formula is as follows:
where c and ξ represent positive numbers greater than 0,/represents the division of the corresponding element between the two tensors;
the augmented lagrangian equation for the rewritten objective function is as follows:
wherein, the first and the second end of the pipe are connected with each other,and &>Represents a lagrange multiplier, μ represents a non-negative penalty factor, a-represents a hadamard product,<·>expressing inner product operation, | · Lixiao F Representing the Frobenius norm.
Preferably, the inputting of the third-order tensor and the prior information weight tensor into the objective function in step 3, and solving the objective function by using the ADMM includes the following steps:
step b1: third order tensor to be constructed by original imageInputting an objective function to be solved;
step b2: initializing parameters of an augmented Lagrange equation, and enabling the iteration number k =0 and the maximum iteration number to be kmax;
wherein S is τ (. Represents a soft threshold shrink operator, S τ (x)=sgn(x)max(|x|-τ,0);
wherein the content of the first and second substances, is->Singular value decomposition of->Is an f-diagonal tensor, and->Here, is greater or less>And &>Respectively denote->And &>Result from a discrete Fourier transform in a third dimension, based on the comparison of the result and the result>Indicating that in the kth iteration>The ith singular value of the ith front slice;
To solve the above equation, the above equation can be solved by dividing it into q sub-problems, which are as follows:
this problem can be solved by a Fast Iterative threshold Shrinkage Algorithm (FISTA) Algorithm (Fast Iterative Shrinkage-Thresholding Algorithm);
step b8: updating mu k+1 =ρμ k Wherein rho represents a growth coefficient, and rho is more than or equal to 1;
step b9: the iteration number k = k +1;
step b10: judging whether k is larger than k max If yes, stopping iteration and turning to the step b11; if not, stopping iteration when the following conditions are met, and turning to the step b11:
if the iteration stop condition is not met and the iteration times are not the maximum value, turning to the step b3;
Preferably, the specific steps of step 4 are: background tensor for inputTake out in sequence>Q frontal slices->And in turn reconstruct the captured background map>For an input target tensor->Take out in order>Q frontal slices->And in turn reconstruct the acquired target map>
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. according to the method, the importance degrees of singular values with different sizes are different, the non-convex logDet function is adopted to constrain tensor ranks, different weights are given to different singular values, and the non-convex property of the function is added, so that the method is closer to the true rank of the tensor, the problem that the target detection accuracy is low due to the fact that the local optimal solution is caused by the fact that the nuclear norm with the same weight is given to the singular values in the existing method is solved, and the target detection and background suppression capabilities are improved;
2. because the actual scene is complicated and changeable, the ATV is introduced to specially restrict the background edge, the ATV can describe the internal smoothness and the definition of the background and is closely related to the gradient (high-frequency component) of the image, the edge is the high-frequency component, the edge in the background can be better described by introducing the ATV regular term, the sparse edge component in the target component is inhibited, the problem that the existing method is difficult to inhibit the highlight edge with sparse property, so that the separated target contains edge noise is solved, and the detection capability in the non-smooth and non-uniform scenes is improved;
3. the invention simultaneously utilizes prior information related to the background and the target, increases the constraint capacity of the target by utilizing the weight of the more prominent target, and simultaneously adds the prior as a part of the regular term into the target function, so that the optimal condition of the target function is stronger, the range of feasible regions is reduced, the speed of achieving the optimal solution is improved, the convergence speed of the algorithm is accelerated, and the robustness of the algorithm is improved;
4. the method converts the traditional infrared small and weak target detection problem into the solution tensor recovery problem, can adaptively separate the target and the background without extracting any feature, and has wider applicability.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of the present invention;
FIG. 2 is an infrared image of the present invention containing a small and weak target;
FIG. 3 is a diagram of the third order tensors constructed from FIG. 2 according to the present invention;
FIG. 4 is the prior information map and the prior information tensor calculated from FIG. 2 according to the present invention;
FIG. 5 is the object tensors of the present invention isolated from FIG. 3;
FIG. 6 is the background tensors of the present invention isolated from FIG. 3;
FIG. 7 is a target image and a background image reconstructed from FIGS. 5 and 6 in accordance with the present invention;
FIG. 8 is a gray scale three-dimensional distribution diagram of the target image of FIGS. 2 and 5 according to the present invention;
FIG. 9 is a diagram illustrating the adaptive threshold segmentation of the target image of FIG. 5 to obtain a detection result according to the present invention;
FIG. 10 is a graph of the detection result and three-dimensional gray scale of FIG. 2 by the LoG method;
FIG. 11 is a graph of the results of the RLCM assay of FIG. 2 and a three-dimensional gray scale;
FIG. 12 is a graph of the result of the IPI method versus the test of FIG. 2 and a three-dimensional gray scale map;
FIG. 13 is a graph of the NIPPS method versus the test results of FIG. 2 and a three-dimensional gray scale;
FIG. 14 is a graph of the results of the RIPT process on the test of FIG. 2 and a three-dimensional gray scale;
FIG. 15 is a schematic diagram of the RIPT method and prior information of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
The features and properties of the present invention are described in further detail below with reference to examples.
Example 1
As shown in fig. 1 to 15, a tensor recovery infrared weak and small target detection method combined with ATV constraint includes the following steps:
step 1: constructing a third-order tensor of the original image;
and 2, step: extracting prior information of an original image, and constructing a prior information weight tensor;
and step 3: using tensor logDet function and tensor l 1 Norm, combining with ATV constraint, constructing an objective function, inputting a third-order tensor and a priori information weight tensor into the objective function, and solving the objective function by using ADMM to obtain a background tensor and a targetA scalar quantity;
and 4, step 4: reconstructing a background image and a target image according to the background tensor and the target tensor;
and 5: and performing self-adaptive threshold segmentation on the target image to determine the position of the target and outputting a target detection result.
In order to improve the accuracy of weak and small target detection, the local optimality problem caused by nuclear norm and the influence of background edges and noise in an image on detection need to be overcome, a tensor logDet function is used as a regular term of a low-rank tensor, and tensor l 1 The norm is used as a regular term of the sparse tensor to construct a tensor recovery target function, different weights are given to different singular values, the non-convexity of the function is added, the tensor recovery target function is closer to the real rank of the tensor, and the problem that the local optimal solution is caused by the adoption of the nuclear norm which is given to the singular values and has the same weight in the existing method is solved; the edge in the background is better described by combining with the ATV, and sparse edge components in the target component are suppressed; the target detection and background suppression capability is improved, and the infrared small and weak target detection accuracy is improved.
Effect analysis was performed according to the attached figures: FIG. 2 shows an infrared image with a complex background, with a very bright white false alarm source in addition to a small target; FIG. 3 is a third order tensor constructed from an original image via step 1FIG. 4 is the prior information extracted by step 2 and the corresponding prior information weight tensor ≦>FIG. 5 shows the background tensor determined by the separation in step 3>And a target tensor +>Fig. 6 is the background image B and the target image T reconstructed by step 4; fig. 7 is a three-dimensional distribution of the gray levels of the original image D corresponding to the target image T, and it can be seen that,the background of the separated target image is well suppressed, and the gray levels of the background at other positions are 0 after the small target is removed; FIG. 8 is the final test result; fig. 9-14 show the results of detection (without threshold segmentation) of the small target in fig. 2 by several other methods (LoG, RLCM, IPI, NIPPS, and RIPT in sequence) and the corresponding gray scale three-dimensional distribution maps, and it can be seen that both LoG and IPI (fig. 10 and 12) are extremely sensitive to background edges and noise, and in RLCM and NIPPS (fig. 11 and 13), in addition to the target, much noise remains, the false alarm rate is high, the target is shrunk to different degrees, and RIPT (fig. 14) does not detect the target. In conclusion, the method and the device have the advantages of strong background suppression capability, extremely small noise, no distortion, excellent target detection effect and greatly improved target detection accuracy.
Example 2
Based on embodiment 1, the steps of the present application are detailed, and technical means for solving the technical problems are described in detail: using the tensor logDet function and the tensor l 1 And norm, combining with ATV constraint, constructing a target function, inputting the three-order tensor and the prior information weight tensor into the target function, and solving the target function by using ADMM to obtain a background tensor and a target tensor.
The step 1 comprises the following steps:
step 1.1: obtaining the infrared image D epsilon R to be processed m×n 245 × 326 in size;
step 1.2: traversing the original image D by adopting a 40 multiplied by 40 sliding window w and 40 step length, and taking a matrix with the size of 40 multiplied by 40 in the sliding window w each time as a front section;
step 1.3: repeating the step 1.2 according to the sliding times of the window (63 in the embodiment) until the traversal is completed, and forming new third-order tensors by all the front slices
As shown in fig. 2, an infrared image with a complex background is shown, and besides a weak and small target, a white false alarm source with high brightness is also shown; as shown in fig. 3, the third-order tensor constructed from the original image through step 1 is represented
Step 2: extracting prior information of an original image, and constructing a prior information weight tensor;
the step 3 comprises the following steps:
step 3.1: using the tensor logDet function, tensor l 1 Norm, combining with ATV constraint, constructing an objective function;
step 3.2: tensor of third orderAnd an a priori information weight tensor>Inputting an objective function, solving the objective function by using ADMM, and solving the background tensor +>And the target tensor pick>
Step 3.1 comprises the following steps:
step 3.1.1: third order tensorComprising a low rank tensor pick>And sparse tensor pick>For separating low rank tensor>And sparse tensor>Will tensorlogDet function as regularization term of low rank tensor, tensor l 1 The norm is used as a regularization term of the sparse tensor, and an objective function is constructed by combining ATV constraint, wherein the formula is as follows:
where η > 0 represents a very small regularization constant, λ and β represent equilibrium coefficients, logDet (-) represents the logDet function of the tensor, and hasMeans for>Represents->I (1. Ltoreq. I. Ltoreq.40) singular values>Is/>The ith frontal slice of the tensor obtained by discrete fourier transform along the third dimension, where l below denotes the same meaning), i | · | | survival 1 The representation tensor l 1 Norm (i.e., sum of singular values of all elements in the tensor), | ·| luminance HTV Represents an anisotropic total variation constraint, which is defined as->Wherein-> And &>Two-dimensional finite difference operators which respectively represent the horizontal and the vertical;
Where c and ξ represent positive numbers greater than 0, the resulting weight tensorIs defined as follows:
where,/represents the division of the corresponding element between the two tensors;
step 3.1.3: introduction of substitute variablesThe original objective function is rewritten as follows: />
The augmented lagrangian equation for the rewritten objective function is as follows:
wherein, the first and the second end of the pipe are connected with each other,and &>Represents a lagrange multiplier, μ represents a non-negative penalty factor, a-represents a hadamard product,<·>expressing inner product operation, | · Lixiao F Representing the Frobenius norm.
Step 3.2 comprises the following steps:
step 3.2.1: third order tensor to be constructed by original imageInputting an objective function to be solved;
step 3.2.2: initializing parameters of an augmented Lagrange equation, and enabling the iteration number to be k =0 and the maximum iteration number to be k max =500,ρ=1.05,μ 0 =0.001,c=1,ξ=0.01,η=0.2,β=0.5;
Step 3.2.3: in the (k + 1) th iteration, fixUpdating a device>The calculation formula is as follows:
wherein S is τ () Representing a soft threshold shrinkage operator, S τ (x)=sgn(x)max(|x|-τ,0);
wherein, the first and the second end of the pipe are connected with each other, is->Singular value decomposition of->Is an f-diagonal tensor, and->Here, is greater or less>And &>Respectively represent->And &>Result from a discrete Fourier transform in a third dimension, based on the comparison of the result and the result>Indicating that in the kth iteration>The ith singular value of the ith front slice;
to solve the above equation, the above equation can be solved by dividing it into q sub-problems, which are as follows:
this problem can be solved by a Fast Iterative threshold Shrinkage Algorithm (FISTA) Algorithm;
step 3.2.8: updating mu k+1 =ρμ k Wherein rho represents a growth coefficient, and rho is more than or equal to 1;
step 3.2.9: the iteration number k = k +1;
step 3.2.10: judging whether k is larger than k max If yes, stopping iteration and turning to the step 3.2.11; if not, stopping iteration when the following conditions are met, and going to step 3.2.11:
if the iteration stop condition is not met and the iteration times are not the maximum value, turning to the step 3.2.3;
step 3.2.11: finding out optimal solution and outputting background tensorAnd the target tensor pick>The output sign with x represents the optimal solution, and the solutions of B and T obtained after iterative convergence are the separated target tensor and the background tensor.
The specific steps of the step 4 are as follows: background tensor for inputTake out in sequence>63 positive slices->And in turn reconstruct the captured background map>For an input target tensor->Take out in sequence>In 63 frontal slices->And reconstructing in sequence the acquired target pictures>
The specific steps of the step 5 are as follows: and performing adaptive threshold segmentation on the target image T, wherein a threshold Th = m + c × σ, m represents the mean value of all gray scales in the target image T, σ represents the standard deviation of all gray scales in the target image T, and c =5, completing segmentation and acquiring a target detection result.
As shown in fig. 7, the background image is calculated and processed by the method of the present invention to obtain the final target image, which completely suppresses the background, has no noise and no distortion; the non-convex logDet function with higher approximate low-rank capability than SNN is adopted to constrain the background, the ATV can describe the internal smoothness and the definition of the background and is closely related to the gradient (high-frequency component) of an image, the edge is the high-frequency component, the edge in the background can be better described by introducing the ATV regular term, the sparse edge component in the target component is inhibited, the problem that the existing method is difficult to inhibit the highlight edge with sparse property to cause edge noise in the separated target is solved, the detection capability in the non-smooth and non-uniform scenes is improved, and the target detection accuracy is improved.
Example 3
Based on the embodiment 1, the embodiment refines the step 2, extracts prior information of the original image, constructs a prior information weight tensor, and utilizes the prior information related to the background and the target to ensure that the target is not distorted, so that the convergence rate of the algorithm is increased, and the robustness of the algorithm is also improved.
The step 2 comprises the following steps:
wherein, K ρ A Gaussian kernel representing variance 2, representing a convolution operation, D σ Indicating that the original is gaussian smoothed with a variance of 9,represents a kronecker product, is present>Indicates that a gradient is being taken>Represents D σ The gradient in the x-direction is such that, device for selecting or keeping>Is shown by D σ Gradient in the y-direction, J 11 Replacement->J 12 Substitute K ρ *I x I y ,J 21 Substitute K ρ *I x I y ,J 22 Replacement->
Wherein, an indicates a Hadamard product;
W m =max(λ 1 ,λ 2 );
W p =W cs *W m ;
To W p Normalization was performed as follows:
wherein, w min And w max Respectively represent W p Minimum and maximum values of (d);
step 2.6: according to a normalized prior information matrix W p Constructing a priori information weight tensorThe construction method comprises the following steps: traversing W with a sliding window W of size 40 x 40 p The small image block in the sliding window w is taken as a front section, and after sliding for 63 times, a third-order tensor-based system is formed>
As shown in fig. 15, (a) is a prior information graph obtained from RIPT, and (b) is a prior information graph obtained by the method, it can be found by observing the two graphs that the prior information graph of the present invention only highlights the target, and RIPT highlights the edge of the target while highlighting the edge of the target; therefore, the invention enhances the target constraint capability by utilizing the weight which more highlights the target; the prior information related to the background and the target is utilized, and the prior is used as a part of the regular term and added into the target function, so that the optimal condition of the target function is stronger, the range of a feasible domain is reduced, the optimal solution reaching speed is improved, the algorithm convergence speed is accelerated, and the algorithm robustness is improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (4)
1. A tensor recovery infrared small target detection method combined with ATV constraint is characterized in that: the method comprises the following steps:
step 1: constructing a third-order tensor of the original image;
step 2: extracting prior information of an original image, and constructing a prior information weight tensor;
and 3, step 3: using the tensor logDet function and the tensor l 1 Norm, combining with ATV constraint, constructing an objective function, inputting a third-order tensor and a prior information weight tensor into the objective function, and solving the objective function by using ADMM to obtain a background tensor and an objective tensor;
and 4, step 4: reconstructing a background image and a target image according to the background tensor and the target tensor;
and 5: performing self-adaptive threshold segmentation on the target image to determine the position of the target, and outputting a target detection result;
specifically, the constructing the objective function in step 3 includes the following steps:
step a1: third order tensorComprising a low rank tensor pick>And sparse tensor>To separate the low rank tensor B and the sparse tensor T, the tensor logDet function is used as the regularization term of the low rank tensor, tensor l 1 The norm is used as a regularization term of the sparse tensor, and an objective function is constructed by combining ATV constraint, wherein the formula is as follows:
min logDet(B,η)+λ||W⊙T|| 1 +β||B|| HTV
s.t.D=B+T
where η > 0 represents a very small regularization constant, λ and β represent equilibrium coefficients, logDet (-) represents the logDet function of the tensor, and hasMeans for>Represents->I (i is more than or equal to 1 and less than or equal to p) th singular value->Is the ith front slice of the tensor obtained by the discrete Fourier transform of B along the third dimension, wherein l below represents the same meaning, | | · | | survival 1 The representation tensor l 1 Norm, | · | luminance HTV Represents an anisotropic total variation constraint, which is defined as->Wherein->D h And D v Two-dimensional finite difference operators which respectively represent the horizontal and the vertical;
step a2: order toRepresenting a sparse weight tensor, defining a weight tensor based on the sparse weight and the prior information weight tensor>The formula is as follows:
W=W s ./W p
where c and ξ represent positive numbers greater than 0,/represents the division of the corresponding element between the two tensors;
step a3: introducing a substitute variable Z = B, and rewriting an original objective function as follows:
min logDet(B,η)+λ||W⊙T|| 1 +β||Z|| HTV
s.t.D=B+T,Z=B
the augmented lagrangian equation for the rewritten objective function is as follows:
wherein, Y 1 And Y 2 Indicating a lagrange multiplier, mu indicating a non-negative penalty factor, an, indicating a hadamard product,<·>expressing inner product operation, | · Lixiao F Represents a Frobenius norm;
in the step 3, the third-order tensor and the prior information weight tensor are input into the objective function, and the step of solving the objective function by using the ADMM comprises the following steps:
step b1: third order tensor to be constructed by original graphInputting an objective function to be solved; />
Step b2: initializing parameters of an augmented Lagrange equation, and enabling the iteration number k =0 and the maximum iteration number to be kmax;
and b3: in the k +1 th iteration, B, Z, Y are fixed 1 ,Y 2 W, update T k+1 The calculation formula is as follows:
wherein S is τ (. Represents a soft threshold shrink operator, S τ (x)=sgn(x)max(|x|-τ,0);
Step b4: fixing T, Z, Y 1 ,Y 2 W, update B k+1 The calculation formula is as follows:
wherein the content of the first and second substances,A=U*S*V T is a singular value decomposition of A, C is an f-diagonal tensor, and ^ s>Here, a combination of>And &>Respectively representing the results of a discrete Fourier transform performed in a third dimension for C and S, based on the first dimension and the second dimension, based on the first dimension>Indicating that in the kth iteration>Epsilon is a constant greater than 0 in order to prevent the denominator from being 0;
and b5: fixing B, T, Y 1 ,Y 2 W, update Z k+1 The following were used:
to solve the above equation, the above equation can be solved by dividing it into q sub-problems, which are as follows:
this problem can be solved by a Fast Iterative threshold Shrinkage Algorithm (FISTA) Algorithm;
step b6: fixing B, T, Z, Y 1 ,Y 2 Update W k+1 The following were used:
W k+1 =W s k+1 ./W p
step b7: fixing B, T, Z, W, updating Y 1 k+1 And Y 2 k+1 The following were used:
Y 1 k+1 =Y 1 k +μ k (T k+1 +B k+1 -D)
Y 2 k+1 =Y 2 k +μ k (B k+1 -Z k+1 )
step b8: updating mu k+1 =ρμ k Wherein rho represents a growth coefficient, and rho is more than or equal to 1;
step b9: the number of iterations k = k +1;
step b10: judging whether k is larger than kmax, if so, stopping iteration, and turning to the step b11; if not, stopping iteration when the following conditions are met, and turning to the step b11:
if the iteration stop condition is not met and the iteration times are not the maximum value, turning to the step b3;
2. The method for detecting the tensor recovery infrared small dim target combined with the ATV constraint as set forth in claim 1, wherein: the step 1 comprises the following steps:
step 1.1: obtaining an original imageWherein m and n represent the length and width of the image, respectively;
step 1.2: traversing an original image D by adopting a sliding window w with the size of P multiplied by P and according to the step length of s;
3. The method for detecting the infrared weak and small target by combining the tensor recovery constrained by the ATV as claimed in claim 1 or 2, which is characterized in that: the step 2 comprises the following steps:
wherein, K ρ A Gaussian kernel function representing the variance ρ representing a convolution operation, D σ Which means that the original image is subjected to a gaussian smoothing filter with variance, sigma, where sigma is greater than 0,represents a kronecker product, -is present>Indicating a gradient is taken over>Represents D σ A gradient in the x-direction, is present>Is shown by D σ Gradient in the y-direction, J 11 ,J 12 ,J 21 And J 22 Are all auxiliary variables;
Wherein, an indicates a Hadamard product;
W m =max(λ 1 ,λ 2 );
W p =W cs *W m ;
To W p Normalization was performed as follows:
wherein w min And w max Respectively represent W p Minimum and maximum values of (d);
step 2.6: according to a normalized prior information matrix W p Constructing a priori information weight tensorThe construction method comprises the following steps: traversing W with a sliding window W of size p p The image small block in the sliding window w is taken as a front section, and after sliding for q times, a third-order tensor, namely prior information weight tensor->
4. The method for detecting the infrared weak and small target by combining the tensor recovery constrained by the ATV as claimed in claim 1 or 2, which is characterized in that: the specific steps of the step 4 are as follows: background tensor for inputQ frontal slices B of B are taken out in order (1) ,B (q) And in turn reconstruct the captured background map>For an input target tensor-> Taking q frontal slices T of T in order (1) ,...,T (q) And in turn reconstruct the acquired target map>/>
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910202771.1A CN109934815B (en) | 2019-03-18 | 2019-03-18 | Tensor recovery infrared small target detection method combined with ATV constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910202771.1A CN109934815B (en) | 2019-03-18 | 2019-03-18 | Tensor recovery infrared small target detection method combined with ATV constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109934815A CN109934815A (en) | 2019-06-25 |
CN109934815B true CN109934815B (en) | 2023-04-14 |
Family
ID=66987341
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910202771.1A Active CN109934815B (en) | 2019-03-18 | 2019-03-18 | Tensor recovery infrared small target detection method combined with ATV constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109934815B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619624B (en) * | 2019-08-22 | 2022-03-01 | 深圳大学 | Image decomposition method and device |
CN110969638B (en) * | 2019-11-12 | 2023-09-29 | 桂林电子科技大学 | Tensor-based background subtraction method and system |
CN110992342B (en) * | 2019-12-05 | 2023-04-18 | 电子科技大学 | SPCP infrared small target detection method based on 3DATV constraint |
CN111191547A (en) * | 2019-12-23 | 2020-05-22 | 中电健康云科技有限公司 | Medical waste online screening method based on hyperspectral deconvolution and unmixing |
CN111738926A (en) * | 2020-01-17 | 2020-10-02 | 西安万宏电子科技有限公司 | Method and system for restoring image |
CN112305548B (en) * | 2020-10-28 | 2022-08-19 | 西北工业大学 | Real-time detection method for moving small target in active sonar echo map |
CN112946636B (en) * | 2021-01-13 | 2023-06-20 | 电子科技大学 | Multi-frequency near-field millimeter wave sparse image reconstruction method |
CN113256585B (en) * | 2021-05-24 | 2022-11-11 | 北京理工大学 | Real-time detection method for small infrared video moving target based on space-time tensor decomposition |
CN113409261B (en) * | 2021-06-13 | 2024-05-14 | 西北工业大学 | Hyperspectral anomaly detection method based on spatial spectrum feature joint constraint |
CN113393402B (en) * | 2021-06-25 | 2022-09-13 | 广东工业大学 | Method for restoring image background based on tensor ring decomposition |
CN113421279B (en) * | 2021-07-07 | 2023-03-31 | 中国科学院光电技术研究所 | Infrared weak and small target detection method based on weighted nuclear norm minimization |
CN113450413B (en) * | 2021-07-19 | 2022-09-27 | 哈尔滨工业大学 | Ship target detection method based on GF4 single-frame image |
CN114742720B (en) * | 2022-03-16 | 2023-03-14 | 哈尔滨工业大学(深圳) | Tensor completion method, tensor completion device, equipment and storage medium |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1169071A (en) * | 1996-02-27 | 1997-12-31 | 佳能株式会社 | Accounting device, communication apparatus, and communication system |
CN1194541A (en) * | 1997-03-25 | 1998-09-30 | 三星电子株式会社 | Design method for NTSC band stop filter and receiver using said filter |
CN101604383A (en) * | 2009-07-24 | 2009-12-16 | 哈尔滨工业大学 | A kind of method for detecting targets at sea based on infrared image |
CN103020897A (en) * | 2012-09-28 | 2013-04-03 | 香港应用科技研究院有限公司 | Device for reconstructing based on super-resolution of multi-block single-frame image, system and method thereof |
CN103605122A (en) * | 2013-12-04 | 2014-02-26 | 西安电子科技大学 | Receiving-transmitting type robust dimensionality-reducing self-adaptive beam forming method of coherent MIMO (Multiple Input Multiple Output) radar |
CN105913448A (en) * | 2016-05-25 | 2016-08-31 | 哈尔滨工业大学 | Hyperspectral image target detection method based on tensor matched subspace |
CN106097321A (en) * | 2016-06-06 | 2016-11-09 | 哈尔滨工业大学 | A kind of polarization high spectrum image object detection method based on tensor representation |
CN106250873A (en) * | 2016-08-16 | 2016-12-21 | 四川大学 | Extracting of Moving Object and device |
CN106610483A (en) * | 2016-11-11 | 2017-05-03 | 长江大学 | MIMO radar angle estimation algorithm based on tensor space and spectral peak search |
CN107038436A (en) * | 2017-05-24 | 2017-08-11 | 哈尔滨工业大学 | A kind of high spectrum image object detection method based on tensor Spectral match filter |
CN108710851A (en) * | 2018-05-21 | 2018-10-26 | 闽南师范大学 | seismic signal random noise attenuation method, terminal device and storage medium |
CN109215025A (en) * | 2018-09-25 | 2019-01-15 | 电子科技大学 | A kind of method for detecting infrared puniness target approaching minimization based on non-convex order |
CN109325446A (en) * | 2018-09-19 | 2019-02-12 | 电子科技大学 | A kind of method for detecting infrared puniness target based on weighting truncation nuclear norm |
CN109447073A (en) * | 2018-11-08 | 2019-03-08 | 电子科技大学 | A kind of method for detecting infrared puniness target based on tensor Robust Principal Component Analysis |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7471975B2 (en) * | 2004-09-22 | 2008-12-30 | The Research Foundation Of State University Of New York | Optical tensor imaging |
US9082164B2 (en) * | 2008-04-04 | 2015-07-14 | Lina Jamil Karam | Automatic cell migration and proliferation analysis |
-
2019
- 2019-03-18 CN CN201910202771.1A patent/CN109934815B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1169071A (en) * | 1996-02-27 | 1997-12-31 | 佳能株式会社 | Accounting device, communication apparatus, and communication system |
CN1194541A (en) * | 1997-03-25 | 1998-09-30 | 三星电子株式会社 | Design method for NTSC band stop filter and receiver using said filter |
CN101604383A (en) * | 2009-07-24 | 2009-12-16 | 哈尔滨工业大学 | A kind of method for detecting targets at sea based on infrared image |
CN103020897A (en) * | 2012-09-28 | 2013-04-03 | 香港应用科技研究院有限公司 | Device for reconstructing based on super-resolution of multi-block single-frame image, system and method thereof |
CN103605122A (en) * | 2013-12-04 | 2014-02-26 | 西安电子科技大学 | Receiving-transmitting type robust dimensionality-reducing self-adaptive beam forming method of coherent MIMO (Multiple Input Multiple Output) radar |
CN105913448A (en) * | 2016-05-25 | 2016-08-31 | 哈尔滨工业大学 | Hyperspectral image target detection method based on tensor matched subspace |
CN106097321A (en) * | 2016-06-06 | 2016-11-09 | 哈尔滨工业大学 | A kind of polarization high spectrum image object detection method based on tensor representation |
CN106250873A (en) * | 2016-08-16 | 2016-12-21 | 四川大学 | Extracting of Moving Object and device |
CN106610483A (en) * | 2016-11-11 | 2017-05-03 | 长江大学 | MIMO radar angle estimation algorithm based on tensor space and spectral peak search |
CN107038436A (en) * | 2017-05-24 | 2017-08-11 | 哈尔滨工业大学 | A kind of high spectrum image object detection method based on tensor Spectral match filter |
CN108710851A (en) * | 2018-05-21 | 2018-10-26 | 闽南师范大学 | seismic signal random noise attenuation method, terminal device and storage medium |
CN109325446A (en) * | 2018-09-19 | 2019-02-12 | 电子科技大学 | A kind of method for detecting infrared puniness target based on weighting truncation nuclear norm |
CN109215025A (en) * | 2018-09-25 | 2019-01-15 | 电子科技大学 | A kind of method for detecting infrared puniness target approaching minimization based on non-convex order |
CN109447073A (en) * | 2018-11-08 | 2019-03-08 | 电子科技大学 | A kind of method for detecting infrared puniness target based on tensor Robust Principal Component Analysis |
Non-Patent Citations (5)
Title |
---|
Tensor Completion for Estimating Missing Values in Visual Data;J. Liu;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20131231;第208-220页 * |
图像自适应中基于形变控制的细缝剪裁方法;张琳等;《信号处理》;20151225(第12期);第82-87页 * |
基于小波和SVR的红外弱小目标检测方法;荣健等;《西南交通大学学报》;20081015(第05期);第5-10页 * |
多传感器分布式检测系统的跨层优化设计;刘向阳等;《半导体光电》;20160815(第04期);第105-112+116页 * |
面向目标检测的稀疏表示方法研究进展;高仕博等;《电子学报》;20150215(第02期);第114-126页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109934815A (en) | 2019-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109934815B (en) | Tensor recovery infrared small target detection method combined with ATV constraint | |
CN109447073B (en) | Infrared weak and small target detection method based on tensor robust principal component analysis | |
Zhu et al. | Joint bi-layer optimization for single-image rain streak removal | |
Ye et al. | Dynamic texture based smoke detection using Surfacelet transform and HMT model | |
CN109215025B (en) | Infrared weak and small target detection method based on non-convex rank approach minimization | |
CN109325446B (en) | Infrared weak and small target detection method based on weighted truncation nuclear norm | |
Park et al. | Single image haze removal with WLS-based edge-preserving smoothing filter | |
Liang et al. | GIFM: An image restoration method with generalized image formation model for poor visible conditions | |
CN110992342B (en) | SPCP infrared small target detection method based on 3DATV constraint | |
Qi et al. | Research on the image segmentation of icing line based on NSCT and 2-D OSTU | |
Kollem et al. | Image denoising by using modified SGHP algorithm | |
Yang et al. | Detail-aware near infrared and visible fusion with multi-order hyper-Laplacian priors | |
CN109285148B (en) | Infrared weak and small target detection method based on heavily weighted low rank and enhanced sparsity | |
Scharfenberger et al. | Image saliency detection via multi-scale statistical non-redundancy modeling | |
CN107085839B (en) | SAR image speckle reduction method based on texture enhancement and sparse coding | |
Yufeng et al. | Research on SAR image change detection algorithm based on hybrid genetic FCM and image registration | |
Zhang et al. | Clustering-based SAR image denoising by sparse representation with KSVD | |
CN114882231A (en) | Infrared small target detection method based on multi-angle prior analysis | |
Kumar et al. | Key frame extraction algorithm for video abstraction applications in underwater videos | |
Liu et al. | Research on lane line segmentation algorithm based on deeplabv3 | |
Xu et al. | Image sequence fusion and denoising based on 3D shearlet transform | |
Al-Fayadh et al. | CT angiography image segmentation by mean shift algorithm and contour with connected components image | |
Lyu | Research on subway pedestrian detection algorithm based on big data cleaning technology | |
Wang et al. | Detecting saliency in infrared images via multiscale local sparse representation and local contrast measure | |
Rashi et al. | Analysis of different image processing strategies |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |