CN109215025A - A kind of method for detecting infrared puniness target approaching minimization based on non-convex order - Google Patents
A kind of method for detecting infrared puniness target approaching minimization based on non-convex order Download PDFInfo
- Publication number
- CN109215025A CN109215025A CN201811116093.9A CN201811116093A CN109215025A CN 109215025 A CN109215025 A CN 109215025A CN 201811116093 A CN201811116093 A CN 201811116093A CN 109215025 A CN109215025 A CN 109215025A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- infrared
- block
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method for detecting infrared puniness target that minimization is approached based on non-convex order, belong to infrared image processing and object detection field；It includes step 1: constructing infrared block of image using sliding window traversal original image；Step 2: approaching minimization building objective function using non-convex order, after infrared block of image is inputted objective function, solve objective function using augmented vector approach and difference method of convex programming and obtain background block image and object block image；Step 3: according to background block image and object block image reconstruction background image and target image；Step 4: the position that Threshold segmentation determines target being carried out to target image, exports object detection results；The present invention solves existing IPI method since the factors such as strong edge, partial noise and sources for false alarms have the characteristics that sparse, to cause small IR targets detection accuracy rate low problem, has achieved the effect that the influence for inhibiting the sparse characteristic of other factors to Detection accuracy.
Description
Technical field
The invention belongs to infrared image processing and object detection fields, especially a kind of to approach minimization based on non-convex order
Method for detecting infrared puniness target.
Background technique
Infrared imagery technique has the characteristics that untouchable, capture details ability is strong, and not by barriers such as cigarette, mists
Influence the detection of the continuous distant object of realization round the clock；Infrared search and tracking IRST (Infrared search and
Track) system is used widely wherein in the fields such as military, civilian, and small IR targets detection technology is as IRST system
A basic function, infrared reconnaissance, infrared early warning, distant object detection in be of great significance.But due to red
In wave section, the texture of target, structural information lack, while remote, complex background, various clutters influence, infrared target
Often it is in spot or dotted, or even floods in the background, it is extremely difficult that this has resulted in small IR targets detection.
Small IR targets detection technology is divided into two major classes: Detection of Small and dim targets based on single frames and based on multiframe
Detection of Small and dim targets, but since the detection technique based on multiframe needs the motion profile of joint multiframe capture target, row
Except the interference of noise, it is therefore desirable to which great calculation amount and amount of storage apply seldom hardware requirement height in Practical Project.Mesh
Before, commonly the detection method based on single frames is divided into following three classes:
(1) background inhibit: background inhibit class method based in infrared image background uniformity it is assumed that using filter pair
The background of infrared image is predicted, then subtracts background from original image again, is carried out Threshold segmentation finally with this and is detected small and weak mesh
Mark.Max-medium filter, Largest Mean filtering, top cap transformation, two-dimentional least means square etc. belong to the scope of background inhibition.
Although such methods realize it is simple, due to noise and do not meet consistency it is assumed that the method that background inhibits easily is made an uproar
The influence of sound clutter causes the inhibitory effect of the infrared image of most of low signal-to-noise ratio very poor.
(2) vision significance: human visual system HVS (Human Visual System) is related to contrast, vision attention
With three kinds of mechanism of eye movement, be directed to it is most assume in infrared image for contrast mechanisms, target is most significant object.
For example, Difference of Gaussian filter calculates Saliency maps using two different Gaussian filters, and target is detected and known
Not；Method based on local contrast, it is high using the small neighbourhood local contrast comprising target, and the background area for the target not included
The low feature of domain local contrast, by calculating local contrast figure, prominent target inhibits background, achievees the purpose that detection.When
When infrared image meets vision significance hypothesis, the available excellent effect of such methods, still, in practical application scene
Under, this hypothesis is difficult to meet, such as conspicuousness sources for false alarms there are when, erroneous detection problem is difficult to overcome, and causes accuracy rate low.
(3) target background separates: what this kind of methods utilized is the non local autocorrelation and mesh of infrared image background
Target detection problems are converted to optimization problem by target sparsity；It can be subdivided into again based on super complete dictionary, low-rank representation
Method and the method restored based on low-rank background and sparse target.First method needs to be constructed not by Gaussian intensity model in advance
With the super complete dictionary of target size and shape, the process for constructing target dictionary is cumbersome, and testing result is influenced by dictionary, and
If when target size and larger change in shape, Gaussian intensity model will be no longer applicable in；Second method is by block iconic model
The original block image of the available low-rank of IPI (Infrared Patch-Image) model, then by the characteristic of target sparse, lead to
Optimization object function is crossed, while recovering background and target image, finally obtains testing result；Second method excellent,
But there are problems that following two: one, since strong edge, partial noise, sources for false alarms also have the characteristics that sparse, inspection can be reduced
The accuracy rate of survey；Two, since the process of objective function optimization needs iteration, it is difficult to reach real-time.Therefore, it is necessary to a kind of infrared
Detection method of small target can overcome problem above.
Summary of the invention
It is an object of the invention to: the present invention provides a kind of infrared small object inspections that minimization is approached based on non-convex order
Survey method solves existing IPI method since the factors such as strong edge, partial noise and sources for false alarms have the characteristics that sparse, causes red
The low problem of outer Dim targets detection accuracy rate.
The technical solution adopted by the invention is as follows:
A kind of method for detecting infrared puniness target being approached minimization based on non-convex order, is included the following steps:
Step 1: infrared block of image is constructed using sliding window traversal original image；
Step 2: approaching minimization building objective function, after infrared block of image is inputted objective function, benefit using non-convex order
Objective function, which is solved, with augmented vector approach and difference method of convex programming obtains background block image and object block image；
Step 3: according to background block image and object block image reconstruction background image and target image；
Step 4: the position that Threshold segmentation determines target being carried out to target image, exports object detection results.
Preferably, the step 1 includes the following steps:
Step 1.1: obtaining infrared image D ∈ R to be processedm×n；
Step 1.2: using size is s traversal original image D for the sliding window W of p × p, by step-length, each sliding window
The matrix-vector that size is p × p in mouth w is converted into p2× 1 column vector；
Step 1.3: step 1.2 being repeated according to window sliding number q until traversal is completed, by all Column vector groups Cheng Xin's
Matrix, that is, infrared block of image
Preferably, the step 2 includes the following steps:
Step 2.1: infrared block of image of input
Step 2.2: the l that tied ranks minimization is measured, weighted1Norm and l2,1Norm constructs objective function；
Step 2.3: infrared block of imageAfter inputting objective function, mesh is solved using augmented vector approach
Scalar functions export background block imageWith object block image
Preferably, the step 2.2 includes the following steps:
Step 2.2.1: assuming that image X ∈ Rm×nIncluding low-rank ingredient A, sparse ingredient E and high frequency noise content N, mesh is constructed
Scalar functions separate low-rank ingredient A and sparse ingredient E, objective function Equation are as follows:
min||A||γ+λ||E||w,1+β||N||2,1
S.t.X=A+E+N
Wherein, λ and β indicates coefficient of balance, | | g | |γI.e.Indicate pseudonorm, | | g | |w,1I.e.Represent the l of weighting1Norm, | | g | |2,1I.e.Represent l2,1Norm；
Step 2.2.2: using augmentation Lagrange's equation optimization object function, and augmentation Lagrange's equation is as follows:
Wherein, Y indicates Lagrange multiplier, and μ indicates that non-negative penalty factor, w indicate weight coefficient matrix, w=1 ∈ Rm ×n,<g>indicates inner product operation, | | g | |FI.e.Indicate Frobenius norm.
Preferably, the step 2.3 includes the following steps:
Step 2.3.1: by infrared block of imageInput objective function, that is, known image X；
Step 2.3.2: solution objective function is iterated based on augmentation Lagrange's equation and difference method of convex programming and is obtained
Take low-rank matrix i.e. background block imageWith sparse matrix, that is, object block image
Preferably, the step 2.3.2 includes the following steps:
Step 2.3.2.1: initialization augmentation Lagrange's equation parameter enables the number of iterations k=0, maximum number of iterations is
maxk；
Step 2.3.2.2: fixed A, N, Y update Ek+1, calculation formula is as follows:
Wherein, Sτ(g) soft-threshold contraction operator, S are indicatedτ(g)=sgn (x) max (| x |-τ, 0)；
Step 2.3.2.3: fixed E, N, Y update A using difference method of convex programmingk+1, calculation formula is as follows:
Enable M=X-Ek+1-Nk+Yk/μk, it is available using difference method of convex programming:
Ak+1=Udiag { σ*}VT,
Wherein, U and V is the left and right singular matrix of M respectively, and diag indicates diagonal matrix, Indicate f (g) in σkGradient, σMIndicate the singular value of M；
Step 2.3.2.4: fixed A, E, Y update Nk+1It is as follows:
Enable Q=X-Ak+1-Ek+1+Yk/μk, then have:
Wherein, [Nk+1]:,iIndicate Nk+1I-th column；
Step 2.3.2.5: fixed A, E, N update Yk+1It is as follows:
Yk+1=Yk+μ(X-Ak+1-Ek+1-Nk+1)；
Step 2.3.2.6: weight w is updatedk+1:
Wherein, C and εTIt indicates to update constant, C >=1, εT> 0；
Step 2.3.2.7: μ is updatedk+1=ρ μk
Wherein, ρ indicates growth factor, ρ > 1；
Step 2.3.2.8: another the number of iterations k=k+1；
Step 2.3.2.9: judging whether k is greater than maxk, if so, stopping iteration, goes to step 2.3.2.10；If it is not,
Judgement | | D0-Ak+1-Ek+1-Nk+1||2/||D0||2Whether≤ε is true, if so, then stop iteration, goes to step 2.3.2.10,
If not, step 2.3.2.2 is gone to, wherein ε indicates loop termination threshold value；
Step 2.3.2.10: optimal solution A is obtained*, E*, N*, export final background block image
With object block image
In conclusion by adopting the above-described technical solution, the beneficial effects of the present invention are:
1. the present invention approaches minimization method using non-convex order, by introducing non-convex γ norm come to infrared block of image
Low-rank ingredient constrained, introduce the l of weighting1Norm improves the approximate ability to sparse ingredient, and combines l2,1Norm is this
Structural sparse norm enhances inhibition to sparse radio-frequency component (such as strong edge), using augmented vector approach with
And difference method of convex programming solves the optimal value of objective function；Existing IPI method is solved due to strong edge, partial noise and void
The factors such as alert source have the characteristics that sparse, to cause small IR targets detection accuracy rate low problem, have reached inhibition other factors
Influence of the sparse characteristic to Detection accuracy effect；
2. the present invention converts small IR targets detection problem to the Solve problems of objective function, do not have to calculate any spy
Sign can adaptively isolate target and background, can efficiently and accurately detect Weak target, while γ norm, l1Model
Several and l2,1Even if the combination of norm three promotes the anti-noise ability for calculating it there are certain noise, algorithm remains to accurately detect
To target, the accuracy rate of small IR targets detection is further increased；
3. number of the present invention due to reducing singular value decomposition, faster, Riming time of algorithm reduces convergence rate, improve
Real-time.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is flow chart of the invention；
Fig. 2 is the infrared image that a width of the invention contains Weak target；
Fig. 3 is the block image that the present invention is constructed by Fig. 2；
Fig. 4 is the present invention by Fig. 3 background block image isolated and object block image；
Fig. 5 is the target image and background image that the present invention is restored by Fig. 4；
Fig. 6 is the gray scale three-dimensional distribution map of the target image in Fig. 2 of the present invention and Fig. 5；
Fig. 7 obtains testing result through adaptive threshold fuzziness by the target image in Fig. 5 for the present invention；
Fig. 8 is testing result figure and Three-Dimensional Gray figure of the IPI method to Fig. 2；
Fig. 9 is testing result figure and Three-Dimensional Gray figure of the NIPPS method to Fig. 2；
Figure 10 is testing result figure and Three-Dimensional Gray figure of the Top-Hat method to Fig. 2；
Figure 11 is testing result figure and Three-Dimensional Gray figure of the MPCM method to Fig. 2；
Figure 12 is effect contrast figure of the invention；
Figure 13 is the sequence chart that comparison of the invention uses.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention, i.e., described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is logical
The component for the embodiment of the present invention being often described and illustrated herein in the accompanying drawings can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiment of the present invention provided in the accompanying drawings is not intended to limit below claimed
The scope of the present invention, but be merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
It should be noted that the relational terms of term " first " and " second " or the like be used merely to an entity or
Operation is distinguished with another entity or operation, and without necessarily requiring or implying between these entities or operation, there are any
This actual relationship or sequence.Moreover, the terms "include", "comprise" or its any other variant be intended to it is non-exclusive
Property include so that include a series of elements process, method, article or equipment not only include those elements, but also
Further include other elements that are not explicitly listed, or further include for this process, method, article or equipment it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described
There is also other identical elements in the process, method, article or equipment of element.
Technical problem: existing IPI method is solved since the factors such as strong edge, partial noise and sources for false alarms have sparse spy
Point, the problem for causing small IR targets detection accuracy rate low；
Technological means:
A kind of method for detecting infrared puniness target being approached minimization based on non-convex order, is included the following steps:
Step 1: infrared block of image is constructed using sliding window traversal original image；
Step 2: approaching minimization building objective function, after infrared block of image is inputted objective function, benefit using non-convex order
Objective function, which is solved, with augmented vector approach and difference method of convex programming obtains background block image and object block image；
Step 3: according to background block image and object block image reconstruction background image and target image；
Step 4: the position that Threshold segmentation determines target being carried out to target image, exports object detection results.
Step 1 includes the following steps:
Step 1.1: obtaining infrared image D ∈ R to be processedm×n；
Step 1.2: using size is s traversal original image D for the sliding window W of p × p, by step-length, each sliding window
The matrix-vector that size is p × p in mouth w is converted into p2× 1 column vector；
Step 1.3: step 1.2 being repeated according to window sliding number q until traversal is completed, by all Column vector groups Cheng Xin's
Matrix, that is, infrared block of image
Step 2 includes the following steps:
Step 2.1: infrared block of image of input
Step 2.2: the l that tied ranks minimization is measured, weighted1Norm and l2,1Norm constructs objective function；
Step 2.3: infrared block of imageAfter inputting objective function, solved using augmented vector approach
Objective function exports background block imageWith object block image
Step 2.2 includes the following steps:
Step 2.2.1: assuming that image X ∈ Rm×nIncluding low-rank ingredient A, sparse ingredient E and high frequency noise content N, mesh is constructed
Scalar functions separate low-rank ingredient A and sparse ingredient E, objective function Equation are as follows:
min||A||γ+λ||E||w,1+β||N||2,1
S.t.X=A+E+N
Wherein, λ and β indicates coefficient of balance, | | g | |γI.e.Indicate pseudonorm, | | g | |w,1I.e.Represent the l of weighting1Norm, | | g | |2,1I.e.Represent l2,1Norm；
Step 2.2.2: using augmentation Lagrange's equation optimization object function, and augmentation Lagrange's equation is as follows:
Wherein, Y indicates Lagrange multiplier, and μ indicates that non-negative penalty factor, w indicate weight coefficient matrix, w=1 ∈ Rm ×n,<g>indicates inner product operation, | | g | |FI.e.Indicate Frobenius norm.
Step 2.3 includes the following steps:
Step 2.3.1: by infrared block of imageInput objective function, that is, known image X；
Step 2.3.2: solution objective function is iterated based on augmentation Lagrange's equation and difference method of convex programming and is obtained
Take low-rank matrix i.e. background block imageWith sparse matrix, that is, object block image
Step 2.3.2 includes the following steps:
Step 2.3.2.1: initialization augmentation Lagrange's equation parameter enables the number of iterations k=0, maximum number of iterations is
maxk；
Step 2.3.2.2: fixed A, N, Y update Ek+1, calculation formula is as follows:
Wherein, Sτ(g) soft-threshold contraction operator, S are indicatedτ(g)=sgn (x) max (| x |-τ, 0)；
Step 2.3.2.3: fixed E, N, Y update A using difference method of convex programmingk+1, calculation formula is as follows:
Enable M=X-Ek+1-Nk+Yk/μk, it is available using difference method of convex programming:
Ak+1=Udiag { σ*}VT,
Wherein, U and V is the left and right singular matrix of M respectively, and diag indicates diagonal matrix, Indicate f (g) in σkGradient, σMIndicate the singular value of M；
Step 2.3.2.4: fixed A, E, Y update Nk+1It is as follows:
Enable Q=X-Ak+1-Ek+1+Yk/μk, then have:
Wherein, [Nk+1]:,iIndicate Nk+1I-th column；
Step 2.3.2.5: fixed A, E, N update Yk+1It is as follows:
Yk+1=Yk+μ(X-Ak+1-Ek+1-Nk+1)；
Step 2.3.2.6: weight w is updatedk+1:
Wherein, C and εTIt indicates to update constant, C >=1, εT> 0；
Step 2.3.2.7: μ is updatedk+1=ρ μk
Wherein, ρ indicates growth factor, ρ > 1；
Step 2.3.2.8: another the number of iterations k=k+1；
Step 2.3.2.9: judging whether k is greater than maxk, if so, stopping iteration, goes to step 2.3.2.10；If it is not,
Judgement | | D0-Ak+1-Ek+1-Nk+1||2/||D0||2Whether≤ε is true, if so, then stop iteration, goes to step 2.3.2.10,
If not, step 2.3.2.2 is gone to, wherein ε indicates loop termination threshold value；
Step 2.3.2.10: optimal solution A is obtained*, E*, N*, export final background block image
With object block image
Technical effect: the present invention approaches minimization method using non-convex order, by introducing non-convex γ norm come to red
The low-rank ingredient of outer block of image is constrained, and the l of weighting is introduced1Norm improves the approximate ability to sparse ingredient, and combines l2,1
This structural sparse norm of norm enhances the inhibition to sparse radio-frequency component (such as strong edge), utilizes augmentation Lagrange
Multiplier method and difference method of convex programming solve the optimal value of objective function；Existing IPI method is solved due to strong edge, part
The factors such as noise and sources for false alarms have the characteristics that sparse, to cause small IR targets detection accuracy rate low problem, have reached inhibition
The effect of influence of the sparse characteristic of other factors to Detection accuracy.
Feature and performance of the invention are described in further detail with reference to embodiments.
Embodiment 1
As shown in figures 1-13, a kind of method for detecting infrared puniness target approaching minimization based on non-convex order, including it is as follows
Step:
Step 1: infrared block of image is constructed using sliding window traversal original image；
Step 2: approaching minimization building objective function, after infrared block of image is inputted objective function, benefit using non-convex order
Objective function, which is solved, with augmented vector approach and difference method of convex programming obtains background block image and object block image；
Step 3: according to background block image and object block image reconstruction background image and target image；
Step 4: the position that adaptive threshold fuzziness determines target being carried out to target image, exports object detection results.
Step 1 includes the following steps:
Step 1.1: obtaining infrared image D ∈ R to be processedm×n, size is 240 × 320；
Step 1.2: use size for 50 × 50 sliding window W, be 10 traversal original image D by step-length, cunning every time
The matrix-vector that size is 50 × 50 in dynamic window W is converted into 2500 × 1 column vector；
Step 1.3: step 1.2 being repeated according to window sliding number 560 until traversal is completed, by all Column vector groups Cheng Xin
Matrix i.e. 2500 × 560 infrared block of image
Step 2 includes the following steps:
Step 2.1: infrared block of image of input 2500 × 560
Step 2.2: the l that tied ranks minimization is measured, weighted1Norm and l2,1Norm constructs objective function；
Step 2.3: infrared block of imageAfter inputting objective function, mesh is solved using augmented vector approach
Scalar functions export background block imageWith object block image
Step 2.2 includes the following steps:
Step 2.2.1: assuming that image X ∈ Rm×nIncluding low-rank ingredient A, sparse ingredient E and high frequency noise content N, mesh is constructed
Scalar functions separate low-rank ingredient A and sparse ingredient E, objective function Equation are as follows:
min||A||γ+λ||E||w,1+β||N||2,1
S.t.X=A+E+N
Wherein, λ and β indicates coefficient of balance,β=1.8, γ=0.002, | | g | |γI.e.Indicate pseudonorm, | | g | |w,1I.e.Represent the l of weighting1Norm, | | g | |2,1
I.e.Represent l2,1Norm；
Step 2.2.2: using augmentation Lagrange's equation optimization object function, and augmentation Lagrange's equation is as follows:
Wherein, Y indicates Lagrange multiplier, and μ indicates that non-negative penalty factor, μ=90, w indicate weight coefficient matrix, w
=1 ∈ Rm×n,<g>indicates inner product operation, | | g | |FI.e.Indicate Frobenius norm.
Step 2.3 includes the following steps:
Step 2.3.1: by infrared block of imageInput objective function, that is, known image X；
Step 2.3.2: solution objective function is iterated based on augmentation Lagrange's equation and difference method of convex programming and is obtained
Take low-rank matrix i.e. background block imageWith sparse matrix, that is, object block image
Step 2.3.2 includes the following steps:
Step 2.3.2.1: initialization augmentation Lagrange's equation parameter enables the number of iterations k=0, maximum number of iterations is
Maxk, enables A, E, N, Y 0,W=1, μ=90；
Step 2.3.2.2: fixed A, N, Y update Ek+1, calculation formula is as follows:
Wherein, Sτ(g) soft-threshold contraction operator, S are indicatedτ(g)=sgn (x) max (| x |-τ, 0)；
Step 2.3.2.3: fixed E, N, Y update A using difference method of convex programmingk+1, calculation formula is as follows:
Enable M=X-Ek+1-Nk+Yk/μk, it is available using difference method of convex programming:
Ak+1=Udiag { σ*}VT,
Wherein, U and V is the left and right singular matrix of M respectively, and diag indicates diagonal matrix, Indicate f (g) in σkGradient, σMIndicate the singular value of M；
Step 2.3.2.4: fixed A, E, Y update Nk+1It is as follows:
Enable Q=X-Ak+1-Ek+1+Yk/μk, then have:
Wherein, [Nk+1]:,iIndicate Nk+1I-th column；
Step 2.3.2.5: fixed A, E, N update Yk+1It is as follows:
Yk+1=Yk+μ(X-Ak+1-Ek+1-Nk+1)；
Step 2.3.2.6: weight w is updatedk+1:
Wherein, C and εTIt indicates to update constant, C >=1, C=1.2, εT> 0, εT=0.4；
Step 2.3.2.7: μ is updatedk+1=ρ μk,
Wherein, ρ indicates growth factor, ρ > 1, ρ=1.1；
Step 2.3.2.8: another the number of iterations k=k+1；
Step 2.3.2.9: judging whether k is greater than maxk, if so, stopping iteration, goes to step 2.3.2.10；If it is not,
Judgement | | D0-Ak+1-Ek+1-Nk+1||2/||D0||2Whether≤ε is true, if so, then stop iteration, goes to step 2.3.2.10,
If not, step 2.3.2.2 is gone to, wherein ε indicates the threshold value of loop termination, ε=10-7；
Step 2.3.2.10: optimal solution A is obtained*, E*, N*, export final background block image
With object block image
The specific steps of step 3 are as follows: for the background block image of inputTake out B0In each column be reconstructed into
The minor matrix of 50 × 50 sizes, then according to the background image B ∈ R for sequentially successively constituting 240 × 320m×n, equal for multiple fritters
The mode of median filtering is taken in the position for including, and determines the gray value of the position, target image T is in the same way by T0
Reconstruct；
The specific steps of step 4 are as follows: adaptive threshold fuzziness, threshold value Th=m+c* σ, wherein m are carried out to target image T
Indicate the mean value of all gray scales in target image T, σ indicates the standard deviation of all gray scales in target image T, and c is indicated between 1-10
Constant, segmentation complete obtain object detection results.
Carry out effect analysis with reference to the accompanying drawings: what Fig. 2 was indicated is the infrared image of a width background complexity, in addition to Weak target it
Outside, there are also the very high white sources for false alarms of brightness；Fig. 3 is the block image D constructed by step 1 by original image0；Fig. 4 be by
Step 2 is by D0The B of recovery0And T0；Fig. 5 is the background B and target figure T reconstructed by step 3；Fig. 6 is original image D and target
The Three-Dimensional Gray figure of image T, it can be seen that the target image isolated has suppressed background well, removes at Small object, remaining
The gray scale of the background of position is 0；Fig. 7 is final testing result；Fig. 8-Figure 11 be several other methods (be successively IPI,
NIPPS, Top-Hat, MPCM) to the testing result of Fig. 2 Small Target (for purposes of illustration only, carrying out binaryzation to result), with
And corresponding gray scale three-dimensional distribution map, it can be seen that remaining four kinds of method does not completely inhibit background, and exists different degrees of
Noise, this will impact subsequent detection and localization, the original graph of one of sequence for selecting, original graph when Figure 13 is comparison
Gray proces have been carried out, therefore have been in celadon；Figure 12 is invention and comparison of the other four kinds of methods in performance, Ke Yiming
Aobvious to find out that signal to noise ratio gain SCRG and Background suppression factor BSF of the invention are maximum, target detection accuracy of the invention is remote
It is much better than other methods.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (6)
1. a kind of method for detecting infrared puniness target for approaching minimization based on non-convex order, characterized by the following steps:
Step 1: infrared block of image is constructed using sliding window traversal original image；
Step 2: approaching minimization building objective function using non-convex order and utilize increasing after infrared block of image is inputted objective function
Wide method of Lagrange multipliers and difference method of convex programming solve objective function and obtain background block image and object block image；
Step 3: according to background block image and object block image reconstruction background image and target image；
Step 4: the position that Threshold segmentation determines target being carried out to target image, exports object detection results.
2. a kind of method for detecting infrared puniness target for approaching minimization based on non-convex order according to claim 1, special
Sign is: the step 1 includes the following steps:
Step 1.1: obtaining original image to be processed i.e. infrared image D ∈ Rm×n；
Step 1.2: using size is s traversal infrared image D for the sliding window W of p × p, by step-length, each sliding window w
Middle size is that the matrix-vector of p × p is converted into p2× 1 column vector；
Step 1.3: step 1.2 being repeated according to window sliding number q until traversal is completed, by the matrix of all Column vector groups Cheng Xin
That is infrared block of image
3. a kind of method for detecting infrared puniness target that minimization is approached based on non-convex order according to claim 1 or 2,
It is characterized by: the step 2 includes the following steps:
Step 2.1: infrared block of image of input
Step 2.2: the l that tied ranks minimization is measured, weighted1Norm and l2,1Norm constructs objective function；
Step 2.3: infrared block of imageAfter inputting objective function, target letter is solved using augmented vector approach
Number output background block imageWith object block image
4. a kind of method for detecting infrared puniness target for approaching minimization based on non-convex order according to claim 3, special
Sign is: the step 2.2 includes the following steps:
Step 2.2.1: assuming that image X ∈ Rm×nIncluding low-rank ingredient A, sparse ingredient E and high frequency noise content N, target letter is constructed
Number separation low-rank ingredient A and sparse ingredient E, objective function Equation are as follows:
min||A||γ+λ||E||w,1+β||N||2,1
S.t.X=A+E+N
Wherein, λ and β indicates coefficient of balance, | | g | |γI.e.Indicate pseudonorm, | | g | |w,1I.e.Represent the l of weighting1Norm, | | g | |2,1I.e.Represent l2,1Norm；
Step 2.2.2: using augmentation Lagrange's equation optimization object function, and augmentation Lagrange's equation is as follows:
Wherein, Y indicates Lagrange multiplier, and μ indicates that non-negative penalty factor, w indicate weight coefficient matrix, w=1 ∈ Rm×n, < g
> indicate inner product operation, | | g | |FI.e.Indicate Frobenius norm.
5. a kind of method for detecting infrared puniness target for approaching minimization based on non-convex order according to claim 4, special
Sign is: the step 2.3 includes the following steps:
Step 2.3.1: by infrared block of imageInput objective function, that is, known image X；
Step 2.3.2: it is low that solution objective function acquisition is iterated based on augmentation Lagrange's equation and difference method of convex programming
Order matrix, that is, background block imageWith sparse matrix, that is, object block image
6. a kind of method for detecting infrared puniness target for approaching minimization based on non-convex order according to claim 5, special
Sign is: the step 2.3.2 includes the following steps:
Step 2.3.2.1: initialization augmentation Lagrange's equation parameter enables the number of iterations k=0, maximum number of iterations maxk；
Step 2.3.2.2: fixed A, N, Y update Ek+1, calculation formula is as follows:
Wherein, Sτ(g) soft-threshold contraction operator, S are indicatedτ(g)=sgn (x) max (| x |-τ, 0)；
Step 2.3.2.3: fixed E, N, Y update A using difference method of convex programmingk+1, calculation formula is as follows:
Enable M=X-Ek+1-Nk+Yk/μk, it is available using difference method of convex programming:
Ak+1=Udiag { σ*}VT,
Wherein, U and V is the left and right singular matrix of M respectively, and diag indicates diagonal matrix, Indicate f (g) in σkGradient, σMIndicate the singular value of M；
Step 2.3.2.4: fixed A, E, Y update Nk+1It is as follows:
Enable Q=X-Ak+1-Ek+1+Yk/μk, then have:
Wherein, [Nk+1]:,iIndicate Nk+1I-th column；
Step 2.3.2.5: fixed A, E, N update Yk+1It is as follows:
Yk+1=Yk+μ(X-Ak+1-Ek+1-Nk+1)；
Step 2.3.2.6: weight w is updatedk+1:
Wherein, C and εTIt indicates to update constant, C >=1, εT> 0；
Step 2.3.2.7: μ is updatedk+1=ρ μk
Wherein, ρ indicates growth factor, ρ > 1；
Step 2.3.2.8: another the number of iterations k=k+1；
Step 2.3.2.9: judging whether k is greater than maxk, if so, stopping iteration, goes to step 2.3.2.10；If it is not, judgement |
|D0-Ak+1-Ek+1-Nk+1||2/||D0||2Whether≤ε is true, if so, then stop iteration, go to step 2.3.2.10, if not
It sets up, goes to step 2.3.2.2, wherein ε indicates loop termination threshold value；
Step 2.3.2.10: optimal solution A is obtained*, E*, N*, export final background block imageAnd target
Block image
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811116093.9A CN109215025B (en) | 2018-09-25 | 2018-09-25 | Infrared weak and small target detection method based on non-convex rank approach minimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811116093.9A CN109215025B (en) | 2018-09-25 | 2018-09-25 | Infrared weak and small target detection method based on non-convex rank approach minimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109215025A true CN109215025A (en) | 2019-01-15 |
| CN109215025B CN109215025B (en) | 2021-08-10 |
Family
ID=64984888
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811116093.9A Active CN109215025B (en) | 2018-09-25 | 2018-09-25 | Infrared weak and small target detection method based on non-convex rank approach minimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109215025B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109839362A (en) * | 2019-03-20 | 2019-06-04 | 北京中科知草科技有限公司 | IR spectrum quantitative analysis method based on gradual noise-removed technology |
| CN109934815A (en) * | 2019-03-18 | 2019-06-25 | 电子科技大学 | A kind of tensor recovery method for detecting infrared puniness target of combination ATV constraint |
| CN110443209A (en) * | 2019-08-08 | 2019-11-12 | 南京邮电大学 | The infrared small target detection method restored based on matrix |
| CN111160181A (en) * | 2019-12-20 | 2020-05-15 | 西北工业大学 | Small target detection method based on infrared video image |
| CN111967466A (en) * | 2020-07-08 | 2020-11-20 | 北京航空航天大学 | Infrared weak and small target detection method utilizing multidirectional sparse low-rank decomposition |
| CN113421279A (en) * | 2021-07-07 | 2021-09-21 | 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 |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101173987A (en) * | 2007-10-31 | 2008-05-07 | 北京航空航天大学 | Multi-module and multi-target accurate tracking apparatus and method thereof |
| US20100179741A1 (en) * | 2009-01-09 | 2010-07-15 | Robert Bosch Gmbh | Lost target function for adaptive cruise control |
| US20120293663A1 (en) * | 2011-05-16 | 2012-11-22 | Sony Corporation | Device for determining disappearing direction and method thereof, apparatus for video camera calibration and method thereof |
| CN103268607A (en) * | 2013-05-15 | 2013-08-28 | 电子科技大学 | Common object detection method on weak supervision condition |
| CN103456015A (en) * | 2013-09-06 | 2013-12-18 | 电子科技大学 | SAR target detection method based on optimal fractional domain Gabor spectrum features |
| CN104933683A (en) * | 2015-06-09 | 2015-09-23 | 南昌大学 | Non-convex low-rank reconstruction method for rapid magnetic resonance (MR) imaging |
| CN107133930A (en) * | 2017-04-30 | 2017-09-05 | 天津大学 | Ranks missing image fill method with rarefaction representation is rebuild based on low-rank matrix |
| CN107680120A (en) * | 2017-09-05 | 2018-02-09 | 南京理工大学 | Tracking Method of IR Small Target based on rarefaction representation and transfer confined-particle filtering |
-
2018
- 2018-09-25 CN CN201811116093.9A patent/CN109215025B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101173987A (en) * | 2007-10-31 | 2008-05-07 | 北京航空航天大学 | Multi-module and multi-target accurate tracking apparatus and method thereof |
| US20100179741A1 (en) * | 2009-01-09 | 2010-07-15 | Robert Bosch Gmbh | Lost target function for adaptive cruise control |
| US20120293663A1 (en) * | 2011-05-16 | 2012-11-22 | Sony Corporation | Device for determining disappearing direction and method thereof, apparatus for video camera calibration and method thereof |
| CN103268607A (en) * | 2013-05-15 | 2013-08-28 | 电子科技大学 | Common object detection method on weak supervision condition |
| CN103456015A (en) * | 2013-09-06 | 2013-12-18 | 电子科技大学 | SAR target detection method based on optimal fractional domain Gabor spectrum features |
| CN104933683A (en) * | 2015-06-09 | 2015-09-23 | 南昌大学 | Non-convex low-rank reconstruction method for rapid magnetic resonance (MR) imaging |
| CN107133930A (en) * | 2017-04-30 | 2017-09-05 | 天津大学 | Ranks missing image fill method with rarefaction representation is rebuild based on low-rank matrix |
| CN107680120A (en) * | 2017-09-05 | 2018-02-09 | 南京理工大学 | Tracking Method of IR Small Target based on rarefaction representation and transfer confined-particle filtering |
Non-Patent Citations (2)
| Title |
|---|
| 赵丹凤: "基于通用对象估计的目标检测与模糊车牌识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陈勇勇: "基于非凸近似的矩阵秩极小化算法研究及应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934815A (en) * | 2019-03-18 | 2019-06-25 | 电子科技大学 | A kind of tensor recovery method for detecting infrared puniness target of combination ATV constraint |
| CN109934815B (en) * | 2019-03-18 | 2023-04-14 | 电子科技大学 | Tensor recovery infrared small target detection method combined with ATV constraint |
| CN109839362A (en) * | 2019-03-20 | 2019-06-04 | 北京中科知草科技有限公司 | IR spectrum quantitative analysis method based on gradual noise-removed technology |
| CN109839362B (en) * | 2019-03-20 | 2021-08-17 | 北京中科知草科技有限公司 | Infrared spectrum quantitative analysis method based on progressive denoising technology |
| CN110443209A (en) * | 2019-08-08 | 2019-11-12 | 南京邮电大学 | The infrared small target detection method restored based on matrix |
| CN110443209B (en) * | 2019-08-08 | 2022-07-29 | 南京邮电大学 | Infrared small target detection method based on matrix recovery |
| CN111160181A (en) * | 2019-12-20 | 2020-05-15 | 西北工业大学 | Small target detection method based on infrared video image |
| CN111160181B (en) * | 2019-12-20 | 2022-07-05 | 西北工业大学 | Small target detection method based on infrared video image |
| CN111967466A (en) * | 2020-07-08 | 2020-11-20 | 北京航空航天大学 | Infrared weak and small target detection method utilizing multidirectional sparse low-rank decomposition |
| CN111967466B (en) * | 2020-07-08 | 2022-08-16 | 北京航空航天大学 | Infrared weak and small target detection method utilizing multidirectional sparse low-rank decomposition |
| CN113421279A (en) * | 2021-07-07 | 2021-09-21 | 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109215025B (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109215025A (en) | A kind of method for detecting infrared puniness target approaching minimization based on non-convex order | |
| CN109325446A (en) | A kind of method for detecting infrared puniness target based on weighting truncation nuclear norm | |
| CN110007366B (en) | Life searching method and system based on multi-sensor fusion | |
| CN109934815A (en) | A kind of tensor recovery method for detecting infrared puniness target of combination ATV constraint | |
| CN109447073A (en) | A kind of method for detecting infrared puniness target based on tensor Robust Principal Component Analysis | |
| CN112285712B (en) | Method for improving detection precision of coasting ship in SAR image | |
| CN109934178A (en) | A kind of method for detecting infrared puniness target based on Kronecker base rarefaction representation | |
| CN112149591B (en) | SSD-AEFF automatic bridge detection method and system for SAR image | |
| CN110135344B (en) | Infrared dim target detection method based on weighted fixed rank representation | |
| Lipschutz et al. | New methods for horizon line detection in infrared and visible sea images | |
| Zhang et al. | Infrared small target detection via interpatch correlation enhancement and joint local visual saliency prior | |
| Li et al. | A small target detection algorithm in infrared image by combining multi-response fusion and local contrast enhancement | |
| Zhao et al. | An adaptation of CNN for small target detection in the infrared | |
| CN110706208A (en) | Infrared dim target detection method based on tensor mean square minimum error | |
| Jiang et al. | Automatic space debris extraction channel based on large field of view photoelectric detection system | |
| Liu et al. | Single-frame infrared small target detection by high local variance, low-rank and sparse decomposition | |
| Wang et al. | Maritime Small Target Detection Based on Appearance Stability and Depth-Normalized Motion Saliency in Infrared Video With Dense Sunglints | |
| Srivastava et al. | Image Pre-processing Algorithms for Detection of Small/Point Airborne Targets. | |
| Xinlong et al. | Research on infrared small target tracking method | |
| Gao et al. | TVPCF: A spatial and temporal filter for small target detection in IR images | |
| Yin | Image micro-target recognition method based on multi-layer empirical mode decomposition algorithm | |
| Gao et al. | A fast detection method for infrared small targets in complex sea and sky background | |
| Sun et al. | Infrared Small-Target Detection Based on Multi-level Local Contrast Measure | |
| Chen et al. | Detection of roads in SAR images using particle filter | |
| Shaoyi et al. | Spatio-temporal saliency fusion based small infrared moving target detection under sea-sky background |
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 |