CN109523575A - Infrared weak and small target detection method - Google Patents
Infrared weak and small target detection method Download PDFInfo
- Publication number
- CN109523575A CN109523575A CN201811343659.1A CN201811343659A CN109523575A CN 109523575 A CN109523575 A CN 109523575A CN 201811343659 A CN201811343659 A CN 201811343659A CN 109523575 A CN109523575 A CN 109523575A
- Authority
- CN
- China
- Prior art keywords
- target
- image
- formula
- background
- infrared
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 59
- 230000001629 suppression Effects 0.000 claims abstract description 19
- 238000001914 filtration Methods 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims description 51
- 238000005516 engineering process Methods 0.000 claims description 14
- 238000005259 measurement Methods 0.000 claims description 14
- 238000012795 verification Methods 0.000 claims description 9
- 238000009825 accumulation Methods 0.000 claims description 7
- 230000002708 enhancing effect Effects 0.000 claims description 7
- 230000006872 improvement Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 7
- 230000006870 function Effects 0.000 description 23
- 238000012360 testing method Methods 0.000 description 13
- 230000008901 benefit Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000003014 reinforcing effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000002401 inhibitory effect Effects 0.000 description 2
- 230000005764 inhibitory process Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 241000283984 Rodentia Species 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000005693 optoelectronics Effects 0.000 description 1
- VLPFTAMPNXLGLX-UHFFFAOYSA-N trioctanoin Chemical compound CCCCCCCC(=O)OCC(OC(=O)CCCCCCC)COC(=O)CCCCCCC VLPFTAMPNXLGLX-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/269—Analysis of motion using gradient-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention discloses an infrared small and weak target detection method, which comprises the following steps: step one, according to an edge stop function of an infrared image, considering the difference between a small target and a background, and constructing an anisotropic background suppression scheme to weaken background interference; step two, based on the traditional high-order cumulant, considering the spatial domain characteristics, establishing a space-time high-order cumulant scheme by utilizing the motion energy of weak and small targets, and performing enhancement processing on candidate targets in the background suppression result; and step three, introducing a scale space theory, and calculating the diameter of the small target, so that a pipeline filtering scheme is improved to fully eliminate a pseudo target, and the small target is accurately detected.
Description
Technical field
The invention belongs to detection fields, and in particular to small IR targets detection algorithm.
Background technique
Since infrared object tracking is completed by long-range detection, cause the contrast of target compared with noise
It is low, and target it is occupied in whole image pixel it is less so that the detection of target and tracking are more difficult.Therefore, multiple
Under miscellaneous background interference, infrared target how is accurately detected as current challenge and hot spot.
For this purpose, researcher devises a series of Dim targets detection scheme.As Wang Jun et al. utilizes shape in document 1
State filtering and variance evaluation method, to protrude object pixel, and calculate the signal-to-noise ratio (SNR) of each pixel, by pixel in image
SNR's high is marked as object pixel, then carries out block analysis to labeled image, accurately extracts in consecutive image sequence
Object pixel, input of the object pixel that will test out as the target tracking algorism of Hough transform, setting dual threshold realize
Effective tracking of target.This scheme belongs to typical single frame detection method, cannot make full use of the real goal of different frame
Motion information between difference, noise jamming can not be eliminated, make in its testing result that there are false targets.
In order to eliminate noise jamming using the continuity of the motion profile of Weak target, scholars propose multi frame detection
Technology, pipeline filter algorithm are one of more typical technologies, if Qi et al. includes true mesh by extracting in document 2
The region of mark and suspicious object, forms corresponding regional area similarity discrepancy mappings, and introduce pipeline filter and reflect to difference
Inject capable processing, will wherein the false targets such as noise filter out, retain real goal, its algorithm of experiment show have compared with
High correct verification and measurement ratio and lower false alarm rate, corresponding signal-to-noise ratio have reached 79dB, have stablized correct verification and measurement ratio and be maintained at
93.6% or so.The program uses regional area similarity discrepancy mappings and inhibits background, fully considered real goal with
The information difference of background can effectively protrude infrared target.But this scheme only considered single features, make its available information compared with
It is few, it is insufficient to the difference descriptive power between background and target, keep its background rejection ability to be improved.
Wang et al. proposes the detection method of small target based on local peak detection and pipeline filter in document 3,
Based on local peak detection mechanism, background forecast is obtained as a result, to obtain comprising including real goal and suspicious object
Regional area, and adaptive threshold method is introduced, the most of non-targeted peak value of removal is eliminated finally, being based on pipeline filter method
The interference of the suspicious objects such as residual noise, experimental result show, calculator has a good correct detection accuracy, about 91.8%,
And lower misclassification error, about 0.0259.The program combines local peak detection and adaptive threshold fuzziness mechanism is come
False target is eliminated, the intensity distribution feature difference between real goal and background information is taken full advantage of, can inhibit exhausted big portion
Point background information, but the selection of this scheme heavy dependence threshold value.
Although both the above takes full advantage of the fortune of real goal based on the multiframe Detection of Small and dim targets of pipeline filter
Difference between dynamic information, can effectively eliminate noise jamming, and still, the pipe diameter of this technology is a fixed value, and real
On border, due to the influence of relative motion relation or imaging system detecting distance between moving target and observer, so that weak
The size of Small object be it is continually changing, therefore, can only by size be less than pipe diameter target identification come out, lack adapt to
Property, corresponding adjustment cannot be made with the variation of Weak target size, cause its detection accuracy bad.
Wherein, above-mentioned document 1, document 2 and document 3 are as follows:
Document 1: Wang Jun, Jiang Zhi, Sun Huiting small IR targets detection and tracking based on Noise Variance Estimation
[J] optoelectronic laser, 2018,29 (3): 305-318；
Document 2:QI H, MO B, LIU F X.Small infrared target detection utilizing
Local Region Similarity Difference map [J] .Infrared Physics and Technology,
2017,72 (2): 131-139；
Document 3:WANG B, DONG L L, ZHAO M.A small dim infrared maritime target
detection algorithm based on local peak detection and pipeline-filtering[J]
.Proceedings of the SPIE, 2016,87 (2): 1012-1019.
Summary of the invention
In order to solve the above-mentioned technical problem, the invention proposes inhibited based on space-time Higher Order Cumulants and anisotropy background
Method for detecting infrared puniness target.
In order to achieve the above object, technical scheme is as follows:
Method for detecting infrared puniness target, comprising the following steps:
Step 1, background based on anisotropy Edge-stopping function inhibit: according to the Edge-stopping function of infrared image,
Consider the difference of Weak target and background, constructs anisotropy background Restrain measurement；
The targets improvement based on space-time Higher-Order Cumulants: step 2 based on traditional Higher Order Cumulants, considers airspace
Characteristic establishes a kind of space-time Higher Order Cumulants scheme using the kinergety of Weak target；
Step 3, the Dim targets detection based on improved pipeline filter: introducing Scale-space theory, calculates Weak target
Diameter, improve pipeline filter scheme.
The present invention discloses a kind of method for detecting infrared puniness target, and main advantage is at three aspects: (1) according to background area
The gradient value difference in domain and real goal, improves Edge-stopping function, allows to inhibit clutter in infrared image and makes an uproar
The background interferences information such as sound；(2) Spatial characteristic and kinergety for combining Weak target, change traditional Higher Order Cumulants
Into candidate target can be highlighted；(3) scale space is introduced into classical pipeline filter method, is allowed to according to small and weak
The variation of target size and adjust pipe diameter, improve its adaptability, improve detection accuracy.
Based on the above technical solution, following improvement can also be done:
As a preferred option, step 1 specifically:
Anisotropic character not only can also keep the edge details and mutation of background smoothly with Steady Background Light region
Region, anisotropic spread function are shown in formula (1):
Wherein, u is gray level image；It is the gradient of image u；For Edge-stopping function；Div is divergence operator；
Edge-stopping function in formula (1)As shown in formula (2), calculated according to the gradient relation of different directions flat
The sliding factor:
Wherein, k is a constant；
If the gradient of infrared image is f (i, j), the gradient operator of the regional area where real goal is formula (3) institute
Show:
Wherein, Up_Grad, Down_Grad, Left_Grad, Right_Grad are the upper and lower, left of image f (i, j) respectively
With right gradient；
F (i, j) is filtered using filter function, corresponding filter function is shown in formula (4):
Formula (2) is updated in formula (3), then is based on formula (4), 2 minimum parameter Min1 on different directions can be calculated,
The mean value of Min2 obtains background suppression result to complete the filtering operation of red image.
Using the above preferred scheme, effectively enhance Weak target, weaken background area.
As a preferred option, step 2 specifically:
Higher Order Cumulants can effectively accumulate target energy in time-domain and spatial domain, can preferably enhance infrared small and weak
Target, the Higher Order Cumulants model of traditional M frame are shown in formula (5):
CMf=E { F0(x, y, t1)+F0(x, y, t2)+…F0(x, y, tM)} (5)
Wherein, F0(x, y, t1) it is background suppression result；t1=1,2 ... M represent the frame number of infrared image；M is that image is tired
The quantity of product frame；
It is described in the movement of the Weak target of consecutive frame using 12 modes；Preceding 5 frame is horizontal movement, and intermediate 5 frames are vertical
Straight movement, last 2 frame is diagonal motion；
Regardless of direction, target is always constantly moved on consecutive frame neighborhood；Therefore, the energy accumulation of moving target
It is realized by the maximum energy value for the M successive image frame that adds up in movement neighborhood；The kinergety accumulation of Weak target can retouch
It states as shown in formula (6)-(7):
Wherein, TPIt is the motor pattern of Weak target；R is the radius for accumulating window；fp(x, y, tM) it is after background inhibits
Sequence image；P0(x, y, tM) represent the t of 12 kinds of modesMThe maximum value of frame；
Convolution (5)~formula (7), then improved M frame Higher Order Cumulants are shown in formula (8):
CMf=E { P0(x, y, t1)+P0(x, y, t2)+…P0(x, y, tM)} (8)
The contrast of all targets in background suppression result effectively improves.
Using the above preferred scheme, the contrast of all targets in background suppression result effectively improves, portion
The intensity of point interference information is weakened, preferably reservation target strength.
As a preferred option, step 3 specifically:
Based on pipeline filter technology, deceptive information is filtered according to its pipeline structure, rejects the mistake of false target
Journey are as follows:
(3.1) it if there is z frame sequence image within the t time, is required according to user, it is long that corresponding pipeline is set；
(3.2) image in pipeline is carried out to first time again and completes 8 connected component labelings, and using it as candidate's filtering pair
As；
(3.3) using mass center corresponding to the region in above-mentioned steps (3.2) as pipeline center, to calculate the region memory
Suspicious object quantity such as formula (9) shown in:
Wherein, (x0, y0) be candidate region mass center；L is duct length；D is pipe diameter；I is into the infrared of pipeline
Image sequence number；M is the total amount of suspicious object.
Method for detecting infrared puniness target according to claim 4, which is characterized in that in step 3, the diameter of pipeline
Calculating process it is as follows:
(3.3.1) is according to Scale-space theory it is found that arbitrary image can divide according to the Gauss nuclear convolution that scale is σ
Solution, as shown in formula (10):
Wherein, f (x, y) is the enhancing image of Higher-Order Cumulants；L (x, y, σ) is scale image；
(3.3.2) according to L (x, y, σ), based on constructing differential scale space with difference of Gaussian, as shown in formula (11):
D (x, y, σ)=L (x, y, n σ)-L (x, y, σ) (11)
Wherein, σ is scale factor；L (x, y, n σ) is the scale space that scale factor is n σ, for Weak target, σ
Minimum value be set as 0.5, n=1.2；
(3.3.3) is re-introduced into difference of Gaussian algorithm and calculates extreme value spatial point；By comparing the middle layer of 8 neighbor pixels
With its bilevel 9 pixel, the extreme value of all consecutive points of its spatial domain and scale domain is found；Again by these extreme points
It is considered as target pixel points, recording its scale is space coordinate (x, y) corresponding to σ, the diameter of target is calculated with this, such as formula (12)
It is shown:
(3.3.4) executes step (3.3.1)~step (3.3.3) next frame sequence image, is σ by scaleiInstitute is right
Space coordinate (the x answeredi, yi) record, pipe diameter d is updated according to formula (12)i。
Using the above preferred scheme, after improved pipeline filter processing, false target has obtained abundant inhibition, weak
Small object is accurately identified out, and corresponding intensity distribution is more uniform, apparent clutter and steady noise etc. does not occur
Signal.
As a preferred option, infrared small and weak detection method further includes step 4, utilizes signal to noise ratio gain G SCR and background
Inhibiting factor BSF assesses infrared target detection precision, shown in computation model such as following formula (13):
Wherein, S is the signal amplitude of Weak target；C represents the standard deviation of background；Out, in respectively represent output, defeated
The infrared image entered.
Using the above preferred scheme, it can preferably reflect infrared target detection precision.
As a preferred option, in step 4, also quantified using ROCs curve characteristic, the computation model of ROCs curve is
Shown in formula (14):
Wherein, PdFor correct verification and measurement ratio；FaFor false alarm rate；True represents the quantity correctly detected；Tptal is Weak target
Quantity；False represents erroneous detection quantity；N-Total is all destination numbers of infrared image.
Using the above preferred scheme, it can preferably reflect infrared target detection precision.
Detailed description of the invention
Fig. 1 is the flow diagram of method for detecting infrared puniness target provided in an embodiment of the present invention.
Fig. 2 is the test result comparison diagram of different background suppressing method, specifically:
Fig. 2 (a) is initial infrared image；
Fig. 2 (b) is the three-dimensional intensity distribution of Fig. 2 (a)；
Fig. 2 (c) is by the transformed image of Top-Hat；
Fig. 2 (d) is Fig. 2 (c) three-dimensional intensity distribution；
Fig. 2 (e) is the image after unmodified Edge-stopping function prediction result；
Fig. 2 (f) is Fig. 2 (e) three-dimensional intensity distribution；
Fig. 2 (g) is the image after the background suppression result by the method for the present invention；
Fig. 2 (h) is Fig. 2 (g) three-dimensional intensity distribution.
Fig. 3 is the motor pattern figure of Weak target provided in an embodiment of the present invention.
Fig. 4 is the enhancing result figure based on improved higher order cumulants model, specifically:
Fig. 4 (a) is the image after the enhancing result of improved higher order cumulants method；
Fig. 4 (b) is the three-dimensional intensity distribution of Fig. 4 (a).
Fig. 5 is the Dim targets detection result figure based on improved pipeline filter, specifically:
Fig. 5 (a) is filter pipeline pattern diagram；
Fig. 5 (b) is the image after Dim targets detection result；
Fig. 5 (c) is the three-dimensional intensity distribution of Fig. 5 (b).
Fig. 6 is the corresponding Dim targets detection result figure of three kinds of algorithms under cloud layer varying background, specifically:
Fig. 6 (a) is initial infrared image；
Fig. 6 (b) is the image by the method for the present invention；
Fig. 6 (c) is the image by 2 method of document；
Fig. 6 (d) is the image by 3 method of document.
Fig. 7 is the corresponding Dim targets detection result figure of three kinds of algorithms under complex background, specifically:
Fig. 7 (a) is initial infrared image；
Fig. 7 (b) is the image by the method for the present invention；
Fig. 7 (c) is the image by 2 method of document；
Fig. 7 (d) is the image by 3 method of document.
Fig. 8 is the ROC curve test chart of 3 method of the method for the present invention, 2 method of document and document.
Specific embodiment
The preferred embodiment that the invention will now be described in detail with reference to the accompanying drawings.
It is infrared small and weak in some of embodiments of method for detecting infrared puniness target in order to reach the purpose of the present invention
Object detection method includes the following steps, as shown in Figure 1:
Step 1, background based on anisotropy Edge-stopping function inhibit: according to the Edge-stopping function of infrared image,
Consider the difference of Weak target and background, constructs anisotropy background Restrain measurement；
The targets improvement based on space-time Higher-Order Cumulants: step 2 based on traditional Higher Order Cumulants, considers airspace
Characteristic establishes a kind of space-time Higher Order Cumulants scheme using the kinergety of Weak target；
Step 3, the Dim targets detection based on improved pipeline filter: introducing Scale-space theory, calculates Weak target
Diameter, improve pipeline filter scheme.
In order to improve the detection accuracy of Weak target, the invention discloses one kind to be based on space-time Higher Order Cumulants with each to different
Property background inhibit method for detecting infrared puniness target, main advantage three aspect:
(1) according to the gradient value difference of background area and real goal, Edge-stopping function is improved, allows to inhibit
The background interferences information such as clutter and noise in infrared image；
(2) Spatial characteristic and kinergety for combining Weak target, improve traditional Higher Order Cumulants, can be highlighted
Candidate target out；
(3) scale space is introduced into classical pipeline filter method, allows to the change according to Weak target size
Change and adjust pipe diameter, improve its adaptability, improves detection accuracy.
In order to further optimize implementation result of the invention, in other embodiment, remaining feature technology phase
Together, the difference is that, in step 1, anisotropic character not only can be smoothly with Steady Background Light region, but also can be with
Keep edge details and the sudden change region of background.Anisotropic spread function are as follows:
Wherein, u is gray level image；It is the gradient of image u；For Edge-stopping function；Div is divergence operator.
Edge-stopping function in formula (1)Mainly calculated according to the gradient relation of different directions it is smooth because
Son:
Wherein, k is a constant.
In infrared image, the gradient of flat site is smaller, thereforeBe worth it is larger, need to this region carry out it is more smooth
Processing；And for sudden change region, gradient is larger, so thatValue is smaller, then without being smoothed to it.Therefore,
For region different in infrared image, need according on different directions target area and other regions feature difference come into
Row processing.If the gradient of infrared image is f (i, j), the gradient operator of the regional area where real goal are as follows:
Wherein, Up_Grad, Down_Grad, Left_Grad, Right_Grad are the upper and lower, left of image f (i, j) respectively
With right gradient.
If the corresponding mean value of two minimum parameters Minl, Min2 using initial anisotropy Edge-stopping function is come
F (i, j) is filtered, then the parameter value in Steady Background Light region is larger, and the parameter value in Weak target region is smaller, makes
It is difficult to effectively enhance target.Corresponding filter function are as follows:
Wherein, " not " is logical operation of negating.
The gradient value in the Steady Background Light region on different directions is smaller, and the gradient value in Weak target region is larger, then its
Corresponding Edge-stopping functionValue becomes smaller, at this point, will lead to if using it to the filtering processing for completing infrared image
Value becomes smaller.Therefore, in order to effectively enhance Weak target, weaken background area, the present invention to Edge-stopping function (formula (15)) into
Row improves, sufficiently to inhibit background interference information:
Finally, formula (2) is updated in formula (3), then formula (4) is based on, 2 minimum parameters on different directions can be calculated
The mean value of Min1, Min2 obtain background suppression result to complete the filtering operation of red image.
In order to embody the advantage of proposed background suppression method, the present invention with classical Top-Hat transformation, it is unmodified respectively to
Anisotropic Edge-stopping function is that control group, using three come filter background interference information, is as a result shown in Fig. 2 (c) by taking Fig. 2 (a) as an example
~2 (h).Wherein, the intensity distribution of initial infrared image is shown in Fig. 2 (b), and there are the interference informations such as more clutter and noise.
According to figure it is found that Top-Hat transformation is bad with the background inhibitory effect of unmodified anisotropy Edge-stopping function, there are more
Deceptive information；And can preferably inhibit the interference information in background using improved anisotropy Edge-stopping function,
Corresponding intensity distribution is more uniform, and Fig. 2 (g)~(h) is shown in distribution.
Step 2 is described in detail below.
Higher Order Cumulants can effectively accumulate target energy in time-domain and spatial domain, can preferably enhance infrared small and weak
Target.The Higher Order Cumulants model of traditional M frame are as follows:
CMf=E { F0(x, y, t1)+F0(x, y, t2)+…F0(x, y, tM)} (5)
Wherein, F0(x, y, t1) it is background suppression result；t1=1,2 ... M represent the frame number of infrared image；M is that image is tired
The quantity of product frame.
But traditional Higher Order Cumulants model only considered the time domain specification of infrared image, have ignored the airspace of signal
Feature keeps its reinforcing effect bad.In this regard, the present invention combines time domain and spatial feature to improve the reinforcing effect of target.In phase
The movement of the Weak target of adjacent frame can be described using 12 modes in Fig. 3.Preceding 5 frame is horizontal movement, and intermediate 5 frames are vertically to transport
Dynamic, last 2 frame is diagonal motion.But regardless of direction, target is always constantly moved on consecutive frame neighborhood.Cause
This, the energy accumulation of moving target can be realized by the maximum energy value for the M successive image frame that adds up in movement neighborhood.It is weak
The kinergety accumulation of Small object can be described as:
Wherein, TPIt is the motor pattern of Weak target, sees Fig. 3；R is the radius for accumulating window；fp(x, y, tM) it is background suppression
Sequence image after system；P0(x, y, tM) represent the t of 12 kinds of modes in Fig. 3MThe maximum value of frame.
Convolution (5)~formula (7), then improved M frame Higher Order Cumulants are as follows:
CMf=E { P0(x, y, t1)+P0(x, y, t2)+…P0(x, y, tM)} (8)
By taking Fig. 2 (g) as an example, after handling using formula (8) it, Fig. 4 is as a result seen.According to figure it is found that in background suppression result
The contrast of all targets effectively improves, and the intensity of part interference information is weakened, preferably reservation target strength, sees
Fig. 4 (b).
Step 3 is described in detail below.
In Fig. 4 (a), however it remains a small amount of background pixel and fixed very noisy, single-frame images detection are difficult to filter
These false targets are based on pipeline in this regard, the present invention realizes this purpose according to the difference between the motion information of real goal
Filtering technique filters deceptive information according to its pipeline structure, sees Fig. 5 (a).Its process for rejecting false target are as follows:
(3.1) it if there is z frame sequence image within the t time, is required according to user, it is long that corresponding pipeline is set.
(3.2) image in pipeline is carried out to first time again and completes 8 connected component labelings, and using it as candidate's filtering pair
As；
(3.3) existing in the region to calculate using mass center corresponding to the region in step (3.2) as pipeline center
Suspicious object quantity:
Wherein, (x0, y0) be candidate region mass center；L is duct length；D is pipe diameter；I is into the infrared of pipeline
Image sequence number；M is the total amount of suspicious object.
Pipeline filter can reject the information such as fixation very noisy in infrared background, and true small and weak mesh is recognized accurately
Mark detection.But the diameter of pipeline is a fixed value, go out it can only by the target identification that size is less than pipe diameter
Come, lacks adaptability, cannot be adjusted accordingly with the variation of Weak target size.For this purpose, present invention introduces scale spaces
Theory calculates pipe diameter.Its process is as follows:
(3.3.1) is according to Scale-space theory[12]It is found that the Gauss nuclear convolution that arbitrary image can be σ according to scale
To decompose:
Wherein, f (x, y) is the enhancing image of Higher-Order Cumulants；L (x, y, σ) is scale image.
(3.3.2) according to L (x, y, σ), based on constructing differential scale space with difference of Gaussian:
D (x, y, σ)=L (x, y, n σ)-L (x, y, σ) (11)
Wherein, σ is scale factor；L (x, y, n σ) is the scale space that scale factor is n σ, for Weak target, σ
Minimum value be set as 0.5, n=1.2.
(3.3.3) is re-introduced into difference of Gaussian algorithm and calculates extreme value spatial point.By comparing the middle layer of 8 neighbor pixels
With its bilevel 9 pixel, the extreme value of all consecutive points of its spatial domain and scale domain can be found.Again by these poles
Value point is considered as target pixel points, and recording its scale is space coordinate (x, y) corresponding to σ, and the diameter of target is calculated with this:
(3.3.4) executes step (3.3.1)~step (3.3.3) next frame sequence image, is σ by scaleiInstitute is right
Space coordinate (the x answeredi, yi) record, pipe diameter d is updated according to formula (12)i。
By taking Fig. 4 (a) as an example, it is completed accurately to detect using the improved pipeline filter of the present invention, as a result sees Fig. 5 (b).By
Figure is it is found that after the processing of improved pipeline filter, and false target has obtained abundant inhibition, and Weak target is accurately identified out
Come, corresponding intensity distribution is more uniform, the signals such as apparent clutter and steady noise does not occur, sees Fig. 5 (c).
In order to embody the advantage of method proposed by the invention, it is completed to test using MATLAB software, meanwhile, it will be current
The higher technology of detection accuracy is the method that document 2 and document 3 are previously mentioned respectively as a control group.Key parameter are as follows: n=
1.2, accumulate radius r=0.8, k=1.5 of window.Experiment sample is Fig. 6 (a) and Fig. 7 (a), and evaluation index is signal to noise ratio gain
With ROC curve.
By Fig. 6 (a) it is found that the contrast of infrared small object is lower, and the cloud layer big rise and fall in background, brightness with
Contrast is higher, leads to complete accurately to detect to it have biggish challenge.Equally, based on Fig. 7 (a) it is found that Weak target can
Information is seldom, and contrast is very low, and has noise jamming in infrared background, and by cloud layer, human eye is difficult to know real goal
Not.Then, target is completed to Weak target therein further according to the detection process of 3 algorithm of mentioned technology, document 2 and document to know
Not, output data is as shown in Figure 6, Figure 7.
According to testing result it is found that interfering for complicated infrared background, method proposed by the present invention can be filtered sufficiently
Deceptive information present in background accurately detects true Weak target, sees Fig. 6 (b), Fig. 7 (b)；Document 2, document 3 this
Although two kinds of technologies have also filtered most of interference information in background, there is different number in the testing result of the two
Pseudo- target is shown in Fig. 6 (c)~Fig. 6 (d) and Fig. 7 (c)~Fig. 7 (d).Especially face the detection of Fig. 7 (a), side of the invention
The advantage of method is more obvious, only remains a small amount of false-alarm, still, document 2,3 technology of document testing result in there are more
False target.The reason is that mentioned technology improves the stopping at anisotropy edge using the difference of true Weak target and background
Function can sufficiently inhibit infrared background, and the spatial feature of Weak target signal is dissolved into traditional Higher Order Cumulants side
In case, it is made to combine time domain and Spatial characteristic, the reinforcing effect of candidate target can be greatly improved, further according to the candidate mesh of enhancing
Scale space corresponding to target pixel improves pipeline filter scheme, can be adaptive to the change of Weak target scale
Change, abundant filtration residue false target retains true Weak target, to improve the detection accuracy of the method for the present invention.And it is literary
Offering 2 is to realize Dim targets detection using regional area similarity discrepancy mappings and traditional pipeline filter, although part
Regional Similarity discrepancy mappings take full advantage of the difference of target and background, can inhibit the big portion in background to a certain extent
Divide interference information, still, the pipeline filter diameter used is fixed, fixation very noisy for being difficult to it in filter background etc.
Interference, causes the detection accuracy of its algorithm undesirable.Document 3 is completed by joint local peak detection with pipeline filter small and weak
Target detection, this local peak detection take full advantage of the feature difference of target and background, can preferably inhibit background, but
It is when fainter than background for target signature, to be easy for part background information to be identified as real goal, and the pipeline filter used
Diameter be it is fixed, can only by size be less than pipe diameter target identification come out, lack adaptability.
For the difference between objective evaluation method proposed by the invention and document 2,3 three of document, text of the present invention draws
Enter step 4, introduces signal to noise ratio gain G SCR (Signal to Clutter Rato Gain), Background suppression factor[14]BSF
(Background Suppression Factor) and ROCs (Receiver operating characteristics) are bent
Line characteristic quantifies.Wherein, GSCR and BSF is the efficiency index of evaluation infrared target detection precision, and value is bigger, then table
Show that precision is higher, computation model are as follows:
Wherein, S is the signal amplitude of Weak target；C represents the standard deviation of background；Out, in respectively represent output, defeated
The infrared image entered.
In addition, the computation model of ROCs curve are as follows:
Wherein, PdFor correct verification and measurement ratio；FaFor false alarm rate；True represents the quantity correctly detected；Total is Weak target
Quantity；False represents erroneous detection quantity；N-Total is all destination numbers of infrared image.
With Fig. 6 (a), Fig. 7 (a) for sample, according to the method for the present invention, the testing result of document 2 and document 3, it is based on formula
(13) it obtains corresponding GSCR and BSF value is shown in Table 1.From the data in the table, of the invention to two kinds of complicated infrared background interference
GSCR that method can obtain and BSF value will be significantly higher than document 2, document 3, respectively 5.931,4.325 and 8.233,
3.712；And GSCR and BSF the value outline of document 2 are lower than method of the invention, to Fig. 6 (a), Fig. 7 (a), value is respectively
5.267,3.652 and 7.804,3.179, GSCR and the BSF value of document 3 are minimum, respectively 3.996,2.986, and
5.817、2.473。
GSCR and the BSF test value of 1 distinct methods of table
ROCs curve can preferably embody the Dim targets detection precision under interference in various degree, for this purpose, the present invention is set
Fixed identical Fa, the different infrared background of 100 width is acquired, 50 Weak targets are synthesized therewith, forms one group of test data set,
And detection is realized to it using method of the invention, document 2 and document 3, the ROCs curve of formation is as shown in Figure 8.According to test
Curve is it is found that for different false alarm rate Fa, the corresponding ROCs curve characteristic of mentioned technology is even more ideal, in arbitrary FaUnder,
Higher correct verification and measurement ratio can be obtained, when false alarm rate is 25%, the correct verification and measurement ratio of mentioned algorithm has just reached 97.3%.And
Document 2,3 two kinds of technologies of document correct verification and measurement ratio will slightly below mentioned method, when false alarm rate be 25% when, the two it is correct
Verification and measurement ratio is respectively 92.2%, 59.7%.
In order to improve detection and recognition performance of the infrared small object under complex background, the present invention is traditional by improving
Higher Order Cumulants and building anisotropy background Restrain measurement, propose new infrared target detection method.
Firstly, considering the difference of Weak target and background according to the Edge-stopping function of infrared image, anisotropy is constructed
Background Restrain measurement, to weaken background interference；Subsequently, based on traditional Higher Order Cumulants, considers Spatial characteristic, utilize small and weak mesh
Target kinergety establishes a kind of space-time Higher Order Cumulants scheme, carries out at enhancing to the candidate target in background suppression result
Reason；Finally, introducing Scale-space theory, the diameter of Weak target is calculated, pipeline filter scheme is improved with this, sufficiently to eliminate
Pseudo- target, to accurately detect Weak target.Test data shows: relative to existing infrared target detection scheme, in noise
Under clutter environment, the present invention can accurately identify real goal, have higher signal to noise ratio gain and background inhibit because
Son shows better ROC curve.
The present invention considers the gradient disparities of infrared background and Weak target region, to traditional anisotropy Edge-stopping letter
Improvement is counted up into, filters infrared background from multiple directions, achievees the purpose that inhibit background information；By the movement energy of infrared target
Measure feature is introduced into traditional Higher Order Cumulants scheme, is significantly improved the contrast of candidate target, is sufficiently reserved target strength；
Meanwhile pipeline filter scheme is improved using Gaussian difference scale, so that it is had adaptive characteristic to Weak target size, thus
Improve the detection accuracy of true Weak target.Experimental data demonstrates the validity and superiority of proposed detection algorithm.
For the preferred embodiment of the present invention, it is noted that for those of ordinary skill in the art, do not taking off
Under the premise of from the invention design, various modifications and improvements can be made, and these are all within the scope of protection of the present invention.
Claims (7)
1. method for detecting infrared puniness target, which comprises the following steps:
Step 1, the background based on anisotropy Edge-stopping function inhibit: according to the Edge-stopping function of infrared image, considering
The difference of Weak target and background constructs anisotropy background Restrain measurement；
The targets improvement based on space-time Higher-Order Cumulants: step 2 based on traditional Higher Order Cumulants, considers that airspace is special
Property, a kind of space-time Higher Order Cumulants scheme is established using the kinergety of Weak target；
The Dim targets detection based on improved pipeline filter: step 3 introduces Scale-space theory, calculates the straight of Weak target
Diameter improves pipeline filter scheme.
2. method for detecting infrared puniness target according to claim 1, which is characterized in that the step 1 specifically:
Anisotropic character not only can also keep the edge details and saltation zone of background smoothly with Steady Background Light region
Domain, anisotropic spread function are shown in formula (1):
Wherein, u is gray level image；It is the gradient of image u；For Edge-stopping function；Div is divergence operator；
Edge-stopping function in formula (1)As shown in formula (2), calculated according to the gradient relation of different directions it is smooth because
Son:
Wherein, k is a constant；
If the gradient of infrared image is f (i, j), the gradient operator of the regional area where real goal is shown in formula (3):
Wherein, Up_Grad, Down_Grad, Left_Grad, Right_Grad are the up, down, left and right of image f (i, j) respectively
Gradient；
F (i, j) is filtered using filter function, corresponding filter function is shown in formula (4):
Formula (2) is updated in formula (3), then is based on formula (4), 2 minimum parameter Min1 on different directions can be calculated, Min2's
Mean value obtains background suppression result to complete the filtering operation of red image.
3. method for detecting infrared puniness target according to claim 2, which is characterized in that the step 2 specifically:
Higher Order Cumulants can effectively accumulate target energy in time-domain and spatial domain, can preferably enhance infrared small and weak mesh
Mark, the Higher Order Cumulants model of traditional M frame are shown in formula (5):
CMf=E { F0(x, y, t1)+F0(x, y, t2)+…F0(x, y, tM)} (5)
Wherein, F0(x, y, t1) it is background suppression result；t1=1,2 ... MRepresent the frame number of infrared image；M is the number of image accumulation frame
Amount；
It is described in the movement of the Weak target of consecutive frame using 12 modes；Preceding 5 frame is horizontal movement, and intermediate 5 frames are vertically to transport
Dynamic, last 2 frame is diagonal motion；
Regardless of direction, target is always constantly moved on consecutive frame neighborhood；Therefore, the energy accumulation of moving target passes through
The maximum energy value of M successive image frame of adding up in movement neighborhood is realized；The kinergety accumulation of Weak target can be described as
Shown in formula (6)-(7):
Wherein, TPIt is the motor pattern of Weak target；R is the radius for accumulating window；fp(x, y, tM) be background inhibit after sequence
Image；P0(x, y, tM) represent the t of 12 kinds of modesMThe maximum value of frame；
Convolution (5)~formula (7), then improved M frame Higher Order Cumulants are shown in formula (8):
CMf=E { P0(x, y, t1)+P0(x, y, t2)+…P0(x, y, tM)} (8)
The contrast of all targets in background suppression result effectively improves.
4. method for detecting infrared puniness target according to claim 3, which is characterized in that the step 3 specifically:
Based on pipeline filter technology, deceptive information is filtered according to its pipeline structure, rejects the process of false target are as follows:
(3.1) it if there is z frame sequence image within the t time, is required according to user, it is long that corresponding pipeline is set；
(3.2) image in pipeline is carried out to first time again and completes 8 connected component labelings, and using it as candidate filtering object；
(3.3) existing in the region to calculate using mass center corresponding to the region in above-mentioned steps (3.2) as pipeline center
Shown in suspicious object quantity such as formula (9):
Wherein, (x0, y0) be candidate region mass center；L is duct length；D is pipe diameter；I is the infrared image into pipeline
Sequence number；M is the total amount of suspicious object.
5. method for detecting infrared puniness target according to claim 4, which is characterized in that described in the step 3
The calculating process of the diameter of pipeline is as follows:
(3.3.1) according to Scale-space theory it is found that arbitrary image can be decomposed according to the Gauss nuclear convolution that scale is σ,
As shown in formula (10):
Wherein, f (x, y) is the enhancing image of Higher-Order Cumulants；L (x, y, σ) is scale image；
(3.3.2) according to L (x, y, σ), based on constructing differential scale space with difference of Gaussian, as shown in formula (11):
D (x, y, σ)=L (x, y, n σ)-L (x, y, σ) (11)
Wherein, σ is scale factor；L (x, y, n σ) is the scale space that scale factor is n σ, and for Weak target, σ is most
Small value is set as 0.5, n=1.2；
(3.3.3) is re-introduced into difference of Gaussian algorithm and calculates extreme value spatial point；By comparing the middle layer of 8 neighbor pixels and its
Bilevel 9 pixels, find the extreme value of all consecutive points of its spatial domain and scale domain；These extreme points are considered as again
Target pixel points, recording its scale is space coordinate (x, y) corresponding to σ, the diameter of target is calculated with this, such as formula (12) institute
Show:
(3.3.4) executes step (3.3.1)~step (3.3.3) next frame sequence image, is σ by scaleiCorresponding
Space coordinate (xi, yi) record, pipe diameter d is updated according to formula (12)i。
6. method for detecting infrared puniness target according to claim 1-5, which is characterized in that described infrared small and weak
Detection method further includes step 4, assesses infrared target detection precision using signal to noise ratio gain G SCR and Background suppression factor BSF,
Shown in its computation model such as following formula (13):
Wherein, S is the signal amplitude of Weak target；C represents the standard deviation of background；Out, in respectively represent output, input
Infrared image.
7. method for detecting infrared puniness target according to claim 6, which is characterized in that in the step 4, also utilize
ROCs curve characteristic quantifies, and the computation model of ROCs curve is shown in formula (14):
Wherein, PdFor correct verification and measurement ratio；FaFor false alarm rate；True represents the quantity correctly detected；Total is the number of Weak target
Amount；False represents erroneous detection quantity；N-Total is all destination numbers of infrared image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811343659.1A CN109523575A (en) | 2018-11-12 | 2018-11-12 | Infrared weak and small target detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811343659.1A CN109523575A (en) | 2018-11-12 | 2018-11-12 | Infrared weak and small target detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109523575A true CN109523575A (en) | 2019-03-26 |
Family
ID=65776483
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811343659.1A Pending CN109523575A (en) | 2018-11-12 | 2018-11-12 | Infrared weak and small target detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109523575A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110111283A (en) * | 2019-05-14 | 2019-08-09 | 武汉高德红外股份有限公司 | The reminding method and system of infrared suspected target under a kind of complex background |
| CN110765631A (en) * | 2019-10-31 | 2020-02-07 | 中国人民解放军95859部队 | Effective imaging pixel-based small target judgment method for infrared radiation characteristic measurement |
| CN111027496A (en) * | 2019-12-16 | 2020-04-17 | 电子科技大学 | Infrared dim target detection method based on space-time joint local contrast |
| CN113343758A (en) * | 2021-04-26 | 2021-09-03 | 西安卓越视讯科技有限公司 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
| CN114693732A (en) * | 2022-03-07 | 2022-07-01 | 四川大学华西医院 | Weak and small target detection and tracking method |
| CN115908807A (en) * | 2022-11-24 | 2023-04-04 | 中国科学院国家空间科学中心 | Method, system, computer equipment and medium for quickly detecting weak and small targets |
| CN116645580A (en) * | 2023-06-05 | 2023-08-25 | 北京邮电大学 | Method and device for detecting infrared dim and small targets based on space-time characteristic difference |
| CN117853411A (en) * | 2023-12-01 | 2024-04-09 | 中国科学院国家空间科学中心 | Infrared small target detection method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005069197A1 (en) * | 2003-12-31 | 2005-07-28 | Lockheed Martin Missiles & Fire Control | A method and system for adaptive target detection |
| CN102819740A (en) * | 2012-07-18 | 2012-12-12 | 西北工业大学 | Method for detecting and positioning dim targets of single-frame infrared image |
| CN103761731A (en) * | 2014-01-02 | 2014-04-30 | 河南科技大学 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
| CN106469313A (en) * | 2016-09-30 | 2017-03-01 | 中国科学院光电技术研究所 | Weak and small target detection method for pipe diameter self-adaptive time-space domain filtering |
| CN106780545A (en) * | 2016-11-25 | 2017-05-31 | 中国科学院云南天文台 | A kind of Weak target energy accumulation Enhancement Method of combination time-space domain |
-
2018
- 2018-11-12 CN CN201811343659.1A patent/CN109523575A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005069197A1 (en) * | 2003-12-31 | 2005-07-28 | Lockheed Martin Missiles & Fire Control | A method and system for adaptive target detection |
| CN102819740A (en) * | 2012-07-18 | 2012-12-12 | 西北工业大学 | Method for detecting and positioning dim targets of single-frame infrared image |
| CN103761731A (en) * | 2014-01-02 | 2014-04-30 | 河南科技大学 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
| CN106469313A (en) * | 2016-09-30 | 2017-03-01 | 中国科学院光电技术研究所 | Weak and small target detection method for pipe diameter self-adaptive time-space domain filtering |
| CN106780545A (en) * | 2016-11-25 | 2017-05-31 | 中国科学院云南天文台 | A kind of Weak target energy accumulation Enhancement Method of combination time-space domain |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110111283A (en) * | 2019-05-14 | 2019-08-09 | 武汉高德红外股份有限公司 | The reminding method and system of infrared suspected target under a kind of complex background |
| CN110765631A (en) * | 2019-10-31 | 2020-02-07 | 中国人民解放军95859部队 | Effective imaging pixel-based small target judgment method for infrared radiation characteristic measurement |
| CN110765631B (en) * | 2019-10-31 | 2023-03-14 | 中国人民解放军95859部队 | Effective imaging pixel-based small target judgment method for infrared radiation characteristic measurement |
| CN111027496A (en) * | 2019-12-16 | 2020-04-17 | 电子科技大学 | Infrared dim target detection method based on space-time joint local contrast |
| CN113343758A (en) * | 2021-04-26 | 2021-09-03 | 西安卓越视讯科技有限公司 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
| CN113343758B (en) * | 2021-04-26 | 2022-03-15 | 西安卓越视讯科技有限公司 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
| CN114693732B (en) * | 2022-03-07 | 2022-11-25 | 四川大学华西医院 | Weak and small target detection and tracking method |
| CN114693732A (en) * | 2022-03-07 | 2022-07-01 | 四川大学华西医院 | Weak and small target detection and tracking method |
| CN115908807A (en) * | 2022-11-24 | 2023-04-04 | 中国科学院国家空间科学中心 | Method, system, computer equipment and medium for quickly detecting weak and small targets |
| CN116645580A (en) * | 2023-06-05 | 2023-08-25 | 北京邮电大学 | Method and device for detecting infrared dim and small targets based on space-time characteristic difference |
| CN116645580B (en) * | 2023-06-05 | 2023-11-14 | 北京邮电大学 | Weak and small target detection method and device based on space-time characteristic difference |
| CN117853411A (en) * | 2023-12-01 | 2024-04-09 | 中国科学院国家空间科学中心 | Infrared small target detection method and system |
| CN117853411B (en) * | 2023-12-01 | 2024-07-05 | 中国科学院国家空间科学中心 | Infrared small target detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109523575A (en) | Infrared weak and small target detection method | |
| CN108009628B (en) | Anomaly detection method based on generation countermeasure network | |
| CN105574855B (en) | Infrared small target detection method under cloud background based on template convolution and false alarm rejection | |
| CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
| CN101655914B (en) | Training device, training method and detection method | |
| CN105319537B (en) | Marine radar co-channel interference suppression method based on spatial coherence | |
| CN102609701B (en) | Remote sensing detection method based on optimal scale for high-resolution SAR (synthetic aperture radar) | |
| CN109446917B (en) | Vanishing point detection method based on cascading Hough transform | |
| CN107657209B (en) | Template image registration mechanism based on finger vein image quality | |
| Li et al. | Road lane detection with gabor filters | |
| CN104766079B (en) | A kind of remote method for detecting infrared puniness target | |
| CN104077613A (en) | Crowd density estimation method based on cascaded multilevel convolution neural network | |
| CN104835178A (en) | Low SNR(Signal to Noise Ratio) motion small target tracking and identification method | |
| CN104834915B (en) | A kind of small infrared target detection method under complicated skies background | |
| CN103353989A (en) | SAR image change detection method based on priori, fusion gray level and textural feature | |
| CN105469428B (en) | A kind of detection method of small target based on morphologic filtering and SVD | |
| CN109559273A (en) | A kind of quick joining method towards vehicle base map picture | |
| CN110706208A (en) | Infrared dim target detection method based on tensor mean square minimum error | |
| CN106557740A (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| CN103295221A (en) | Water surface target motion detecting method simulating compound eye visual mechanism and polarization imaging | |
| CN104680536A (en) | Method for detecting SAR image change by utilizing improved non-local average algorithm | |
| CN112329764A (en) | Infrared dim target detection method based on TV-L1 model | |
| CN109828275B (en) | Identification method and system for zero-degree layer bright band | |
| Kang et al. | GuidedMixup: an efficient mixup strategy guided by saliency maps | |
| CN106780545A (en) | A kind of Weak target energy accumulation Enhancement Method of combination time-space domain |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |
|
| RJ01 | Rejection of invention patent application after publication |