CN106508046B - One kind is based on multiple dimensioned bilateral optimized detection method of small target - Google Patents
One kind is based on multiple dimensioned bilateral optimized detection method of small targetInfo
- Publication number
- CN106508046B CN106508046B CN201110016175.8A CN201110016175A CN106508046B CN 106508046 B CN106508046 B CN 106508046B CN 201110016175 A CN201110016175 A CN 201110016175A CN 106508046 B CN106508046 B CN 106508046B
- Authority
- CN
- China
- Prior art keywords
- image
- background
- tau
- target
- infrared image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
One kind is based on multiple dimensioned bilateral optimized infrared image detection method of small target, including：Using the multiple dimensioned multi-direction infrared image for decomposing reading of tower antithesis tree anisotropic filter group, obtain and original image size identical subband；The coefficient value that its high frequency direction subband is changed using bilateral filtering method, the coefficient value for changing its low frequency sub-band using median filter method；Step by step the subband travel direction and yardstick changed are reconstructed using the inverse transformation of tower antithesis tree anisotropic filter group, obtain the background image that estimates, obtain infrared image background histamine result；Image is suppressed to carry out optimization process background using ant colony；Split therefrom extraction Weak target to image.The present invention can not only retain and strengthen Weak target information, be accurately positioned target, while the strong edge information of the complex backgrounds such as the continuous cloud for rising and falling of cloud layer edge, large area preferably can also be suppressed, be effectively improved image entirety signal to noise ratio and contrast.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to suppresses infrared image background clutter and extracts weak
The method of Small object.This method is that tower antithesis tree compound is combined to wave filter group with bilateral filtering theory
And ant group algorithm is used into detection method of small target therein, it is mainly used in infraed early warning system detection
Distant object.
Background technology
In modern war, radar is faced with increasingly serious confrontation and threatened, and passive or passive detection skill
Art is to solve one of effective way that radar electronic warfare is threatened, wherein being one important using passive infrared detection
Research direction, it can Secondary RADAR System effectively improve existence and the counterforce capability of area defence system.
But when target range is far, its imaging area on infrared camera photosurface is very small, and target
Contrast with background is relatively low, often shows as being submerged in complex background (such as strong fluctuating, strong edge or big
The continuous cloud layer of area) in several picture points, as Weak target.If will reliably, stably, exactly
Detect and track this kind of target, then image must be pre-processed, Weak target is miscellaneous from complex background
Extracted in ripple, and high performance small IR targets detection technology is wherein important and crucial one
Item technology.
Nearly two during the last ten years, and Infrared DIM-small Target Image background suppression technology obtains larger development, mainly
There are the methods such as time-domain filtering, spatial domain, frequency domain, wavelet field and theory of partial differential equations.These are common red
Outer target background suppressing method, meets some demands in the field.However, with the development of application,
The Fast Detection Technique of remote small dim moving target, as new active demand.But, work as background
During for complicated structured background, this kind of filtering method can not smooth edges completely, so as to cause to small and weak
The detection probability reduction of moving target, false-alarm probability increase.In this case, in order that useful mesh
Mark feature is retained and effectively strengthened, then the suppression for having to carry out structured background self adaptation will
Weak target is therefrom detected.
The content of the invention
It is an object of the invention to provide a kind of method for detecting infrared puniness target, existed with solving prior art
When background is complicated structured background, the problem of low to the detection probability of small dim moving target.
To solve described problem, it is weak that the present invention provides a kind of infrared image based on multiple dimensioned bilateral optimization
Small target detecting method, including：
(1) infrared image is read in；
(2) background suppression is carried out to the infrared image of reading, the background process of inhibition includes：
A. exploded view picture
The scale parameter and direction number of exploded view picture are set, using tower antithesis tree anisotropic filter group to reading in
Image is decomposed by setting scale parameter and direction initialization number is decomposed, and the picture size sum after decomposition at different levels is upper
The size of first order image, the number sum of the image after decomposition at different levels is the number of upper level image；
B. the high frequency direction sub-band coefficients of all directions are changed using bilateral filtering method；
C. low frequency sub-band coefficient is changed using median filter method；
D. reconstructed image
Using the inverse transformation of tower antithesis tree anisotropic filter group step by step to the reconstruct of sub-band coefficients travel direction and chi
Degree reconstruct, obtains the background image of estimation；
E. target residual image is obtained；
The background image subtraction for the estimation that the infrared image of reading is obtained with reconstruct, is carried on the back
The result images that scape suppresses；
(3) the step of extracting Weak target, the extraction Weak target includes：
A. ant group algorithm is used, the result images after suppressing by the use of background calculate ant as heuristic information
State transition probability；
B. after being circulated throughout every time, pheromones are updated according to the Pheromone update mechanism in ant system；
(4) result images are obtained
According to pheromones distribution segmentation image zooming-out target.
Further, extracting mesh calibration method includes：The path of Ant Search with cycle-index increase gradually
Restrained to target area, its pheromones for showing as leaving on destination path in ant group algorithm is significantly greater than
Other regions, after given number of iterations and cycle-index is reached, if the pheromones on node (i, j) are met
τ″I, j≥τ′max
Then node (i, j) is labeled as target, is otherwise background.
Further, the background histamine result is entered as the heuristic information in ant group algorithm, and to pheromones
Row is secondary to be updated.
Further, the infrared image includes the Weak target for being submerged in strong varying background clutter, described multiple
The cloud that miscellaneous background includes cloud layer edge, large area continuously rises and falls.
Further, Scale Decomposition, every grade of decomposition are carried out to image using tower antithesis tree anisotropic filter group
Include the two-dimentional multiresolutional filter device group and a pair of routines 2 of two passagesnThe two-dimensional directional filter of passage
Ripple device group.
Further, the step of reading in infrared image includes：Infrared image is produced using red phase instrument, using meter
Calculate switch disk and mouse and the infrared image is read in into computer.
Method provided by the present invention is carried out in tower antithesis tree anisotropic filter group transform domain to infrared image
Decompose, the echo signal in the anisotropy region after decomposition is strengthened using bilateral filtering method, and
Echo signal is intactly extracted using the optimization performance of ant group algorithm, target in original image is effectively kept
Radiation characteristic, to the homogeneous region complex background inhibition function admirable of infrared image, makes texture region
Coherent spot effectively removed, texture structure is clear, and every objective evaluation index is better than existing
A variety of methods.
The present invention has advantages below compared with prior art：
First, the present invention carries out multiple dimensioned and multi-direction decomposition using the multiple anisotropic filter of tower antithesis tree,
Because it has the characteristics such as abundant yardstick, directional information and translation invariance so that infrared image is through place
Echo signal enhancing, contour edge information weaken after reason, improve the positioning precision of target.
Second, the present invention is handled after decomposing on the basis of multiple dimensioned multi-direction decomposition using bilateral filtering
High-frequency sub-band coefficient, it is contemplated that the pixel of single isolated positions and the correlation of adjacent pixel and effect,
The information such as regional area gray scale and geometry similitude is taken full advantage of, retains well and strengthens small and weak
Echo signal, is effectively improved the overall signal to noise ratio of image and contrast.
3rd, the present invention seeks the performance of optimum solution using artificial ant algorithm, in image preprocessing
On the basis of, the Weak target after effectively background is suppressed in image is extracted, and precise marking goes out target
Position in the picture.
Brief description of the drawings
Fig. 1 is the detection method of small target provided in an embodiment of the present invention based on multiple dimensioned bilateral optimization
Flow chart；
Fig. 2 is the flow chart that background provided in an embodiment of the present invention suppresses；
Fig. 3 is the flow chart of extraction Weak target provided in an embodiment of the present invention；
Fig. 4 is tower antithesis tree anisotropic filter group structural representation, wherein Fig. 4 a in the embodiment of the present invention
To decompose part, Fig. 4 b are reconstruct part.For multi-scale expression, block P and Q can be iterated lower
Yardstick；
Fig. 5 is double tree anisotropic filter group frequency domain decomposition schematic diagrames in the embodiment of the present invention, and wherein Fig. 5 a are
Two grades of decomposition n=2, Fig. 5 b are that three-level decomposes n=3；
Fig. 6 suppresses the Performance comparision of the complicated cloudy background of texture for the present invention with max-medium filter method
Design sketch, wherein Fig. 6 a are undressed original image, and Fig. 6 b are the place of max-medium filter method
Design sketch is managed, Fig. 6 c suppress the design sketch after background for the present invention, and Fig. 6 d examine for Weak target of the present invention
The design sketch of survey；
Fig. 7 suppresses the Performance comparision of the continuous cloudy background of large area for the present invention with max-medium filter method
Design sketch, wherein Fig. 7 a are undressed original image, and Fig. 7 b are the place of max-medium filter method
Design sketch is managed, Fig. 7 c suppress the design sketch after background for the present invention, and Fig. 7 d examine for Weak target of the present invention
The design sketch of survey；
Fig. 8 suppresses the Performance comparision effect of strong fluctuating cloudy background for the present invention with max-medium filter method
Figure, wherein Fig. 8 a are undressed original image, and Fig. 8 b are imitated for the processing of max-medium filter method
Fruit is schemed, and Fig. 8 c suppress the design sketch after background for the present invention, and Fig. 8 d are Dim targets detection of the present invention
Design sketch.
Embodiment
Below in association with drawings and Examples, the present invention will be further described：
Fig. 1 is the detection method of small target provided in an embodiment of the present invention based on multiple dimensioned bilateral optimization
Flow chart, the detection method of small target based on multiple dimensioned bilateral optimization that the present embodiment is provided includes：
Read in infrared image；Background suppresses；Weak target is extracted；Obtain result images.
Step one, infrared image is read in.
One width infrared image is produced using thermal infrared imager, and will be described using the keyboard and mouse of computer
Infrared image is read in computer.Fig. 6 a, Fig. 7 a, the original image shown in Fig. 8 a are computer equipment
The untreated infrared image read in.The infrared image includes and is submerged in the small and weak of strong varying background clutter
Target, the cloud that the complex background includes cloud layer edge, large area continuously rises and falls.As Fig. 6 a, Fig. 7 a,
Shown in Fig. 8 a, background is complex in untreated infrared image, and Weak target is difficult to be detected.
Step 2, carries out background suppression, to the infrared image of reading by the background clutter in image as far as possible
Ground is removed.
Fig. 2 is the flow chart that background provided in an embodiment of the present invention suppresses, including：To the infrared image
Carry out multi-resolution decomposition, multi-direction decomposition；On the basis of being decomposed to image, max-medium filter
Handle low frequency part, bilateral filtering processing HFS；Then reconstructed image, finally by obtaining target
Residual image obtains background histamine result.
(1) multiple dimensioned, multi-direction exploded view picture
The scale parameter and direction number of the decomposition can be set according to the requirement of accuracy of detection.
Decomposed using tower antithesis tree anisotropic filter group (PDTDFB) by the scale parameter and direction number of setting
The infrared image of reading, the picture size sum after decomposition at different levels is the size of upper level image, each fraction
The number sum of image after solution is the number of upper level image.
In the present embodiment, 3 grades of Scale Decompositions are carried out by tower antithesis tree anisotropic filter group, every grade
Direction number carries out multiple dimensioned and multi-direction separation by 4/4/8 pair of image containing background clutter successively, obtains each chi
Spend the sub-band coefficients of all directions.
Tower antithesis tree anisotropic filter group is by translatable tower wave filter and an antithesis tree 2nThe side of passage
Organize and realize to wave filter (antithesis tree anisotropic filter, DT-DFB).Utilize translatable tower wave filter group
The picture breakdown of a multiresolution is provided, DT-DFB extracts directional information, obtains image in different chis
Decomposition coefficient on degree and different directions.The first layer of tower antithesis tree anisotropic filter group decomposes such as Fig. 4
(a) shown in, infrared image x (n) is read in tower antithesis tree anisotropic filter group, in translatable tower
Pass through high-pass filter R in mode filter group0(w) with low pass filter L0(w) it is two by picture breakdown
Point：HFS and low frequency part.Wherein, HFS is decomposed in different directions, low frequency part
Then enter secondary decomposable process, high and low frequency two parts are decomposed into again；Secondary high-pass filter R1(w)
The radio-frequency component of output is defeated via its original orientation wave filter and antithesis anisotropic filter respectively by DT-DFB
It is born into 2nThe real part and imaginary part of the individual complex valued sub bands represented with plural form,Represent complex valued sub bands
Real part,The imaginary part of complex valued sub bands is represented, i is hierarchical layer series.Low pass filter L1(w) export
Low-frequency component be then again introduced into block P in next layer of sub-resolution decomposition, i.e. Fig. 4 (a) pass through it is continuous
Iteration provides multi-resolution decomposition.
Antithesis tree anisotropic filter group (DT-DFB) can effectively identification image directional information.
DT-DFBs is appreciated that not associated by two and construction process exactly the same direction wave filter (DFBs)
Tree structure is constituted, and corresponds respectively to the real part and imaginary part of complex value image.The structure can be the frequency of signal
Spectrum [- π, π] is divided into 2nIndividual wedge area, n is the direction coefficient of every layer of classification, and each wedge area obtains phase
Answer the detailed information in direction, its decomposing schematic representation as shown in figure 5, wherein Fig. 5 (a) be n=2 when two-stage
Directional Decomposition, is divided into 4 directions by 0 °, 45 °, 135 °, 180 ° (angles with x-axis) by image and enters
Row is decomposed；Fig. 5 (b) be n=3 when three-level Directional Decomposition, by image by 0 °, 22.5 °, 45 °, 67.5 °,
90 °, 112.5 °, 135 °, 157.5 °, 180 ° of points of 8 directions are decomposed.
By the multiple dimensioned multi-direction exploded view as the step of after, infrared image is broken down into high frequency direction
Subband and low frequency sub-band.High frequency direction subband corresponds to the closeer part of image texture, and the part includes mesh
Mark the information such as information and cloud layer edge.Low frequency sub-band corresponds to the thinner part of image texture, the part
Include the relatively flat region of background in image.
(2) the high frequency direction sub-band coefficients of all directions are changed using bilateral filtering method.
Represent after being decomposed using the conversion of tower antithesis tree anisotropic filter group
Each sub-band coefficients, whereinFor low frequency sub-band coefficient,For each band logical sub-band coefficients.In
It is that adaptive background estimation can be expressed as：
In formula, W={ w (m, n) } is weight coefficient, and Ω is some neighborhood of bandpass region (p, q).Its weight coefficient
W is：
In formula, σdAnd σrRespectively control space and sub-band coefficients domain the weights free degree, its respectively with sub-band coefficients
Between distance it is relevant with coefficient value can pass through limited number of time experiment obtain.It is a certain in the equal representative images of m and n
The two-dimensional coordinate of point；Ω () is subbandNeighborhood space, in representative image centered on certain point one
Individual region, the size in the region should be chosen according to target sizes, and generally 3 × 3；N (m) is represented at m points
Subband coefficient values be image gray value；Ch is normaliztion constant.
Because bilateral filtering combines space and the filtering of two kinds of gray scale domain, weight w depends on subband spatial and son
Euclidean distance with coefficient value, therefore the complicated situation of subband background change is adapted to, to stronger in subband
Structural Characteristics background component inhibition it is obvious.
(3) low frequency sub-band coefficient is changed using median filter method.
In the embodiment of the present invention, choose size and be 5 × 5, be centrally located at low frequency sub-band coefficient
The spectral window of position (s, t), the sampled value to 25 sub-band coefficients x (s, t) in the window is carried out from small to large
Sequence, takes the 13rd number in the middle of it as center coefficient x (s, t) filter result, to be shown below：
The various traversal low frequency sub-band more thanEach coefficient, the background subband that can be changedMedium filtering can keep the edge feature of original image, filter spike interference and point-like Small object etc.
High fdrequency component, Weak target that can effectively in the background of place to go estimates background sub-band coefficients.
(4) reconstructed image
Tower antithesis tree anisotropic filter group reconstructed subband coefficient is used in the embodiment of the present invention, background is obtained
Suppress image.Restructuring procedure is the inverse process of tower antithesis tree anisotropic filter component solution.Such as Fig. 4 (b)
It is shown,WithBy being passed through after its original orientation wave filter and the processing of antithesis anisotropic filter group by high pass
Wave filter R1(- w) synthesizes HFS, obtained HFS and low pass filter L1The low frequency of (- w) formation
Low pass filter L partially synthetic and via upper level0(- w) processing obtains the low frequency part of upper level, and with
The HFS of upper level merges again, and reconstructed image y (n) is obtained by constantly repeating the process.Reconstruct
Process is obtained by reconstructed high frequency part and low frequency part, in such as Fig. 4 (b) block Q by continuous iteration come
Reconstruct and synthesize last image.
First, reconstructed high frequency directional subbandRemove directional information,
Obtain high-frequency sub-band coefficientSecondly, reconstructed high frequency subbandAnd low frequency sub-band
Obtain the low frequency sub-band of upper levelThe reconstruct of upper level is similarly carried out again and finally obtains estimation Background
As CB。
(5) target residual image is obtained
The background image of estimation obtained will be reconstructed with original image to subtract each other, i.e., by step one read in it is infrared
Image and the estimation background image C now obtainedBSubtract each other, just can obtain background histamine result.Such as Fig. 6 c,
Shown in Fig. 7 c, Fig. 8 c.
Step 3, extracts Weak target
Fig. 3 is the flow chart of extraction Weak target provided in an embodiment of the present invention, including：At the beginning of ant colony is set
Beginning parameter, set initial parameter includes ant number and cycle-index；Ant colony is randomly dispersed in image
On pixel；Heuristic information is set；Calculate state transition probability；Whether fresh information element, calculating owns
Ant travels through, if it is secondary fresh information number, and calculates whether reach designated cycle number of times, if
Designated cycle number of times is reached, then extracts target；If not reaching designated cycle number of times, calculating is returned to
The step of transfering state probability；If not all ants traversal, also return to and calculate transfering state probability
Step.
(1) state transition probability of ant is calculated；
Assuming that quantity is randomly dispersed in the pixel of image after the background suppression that step 2 is obtained for m ant colony
On, and moved in a pixel is M × N image, each point in pixel is counted as node.
In ant colony search procedure, each ant is pressed according to pheromones and heuristic information from its 8 field node
State transition probability selects next node, until meeting search termination condition, and travels through all ants.
Ant calculates transition probability according to the pheromones and the heuristic information in path on each path, and it is node to make R
The set of all nodes in 8 fields of (i, j), n-th ant is transferred to adjacent node (r, s) from node (i, j)
Probability be：
In formula, α and the weight that β is control information element and heuristic information；τI, jIt is the pheromones on node (i, j)；
μ(i, j)It is the heuristic information on node (i, j).It regard the background histamine result obtained in step 2 as ant colony herein
Heuristic information in algorithm, is conducive to ant preferably to recognize the target area in image, more accurately makes
Ant is obtained to converge to target area.
(2) fresh information element；
After every Ant Search, pheromones are updated according to the Pheromone update mechanism in ant system,
I.e.：
Wherein, when kth ant is transferred to (i, j) place,Other situations are then
The update mechanism of traditional ant group algorithm pheromones can cause the pheromones of the former destination path accessed
It is intended to zero with the increase of cycle-index, all cycle-indexes is traveled through for this and will be believed after each circulation
The codomain scope of breath element is limited in [τmin, τmax] interval interior, then there is second of renewal of pheromones：
Result images after being suppressed in described step three using artificial ant algorithm to background optimize place
Reason, carries out Bio-simulated Evolution the features such as using the robustness of ant algorithm, positive feedback, concurrency.
Step 4, obtains result images
According to pheromones distribution segmentation image zooming-out target.The path of Ant Search with cycle-index increase
Gradually restrained to target area, the pheromones that it shows as leaving on destination path in ant group algorithm are obvious
More than other regions, after given number of iterations and cycle-index is reached, if the pheromones on node (i, j)
Meet
τ″I, j≥τmax
Then node (i, j) is labeled as target, is otherwise background.Thus Weak target in the embodiment of the present invention is just obtained to examine
The result images of survey, as shown in Fig. 6 d, Fig. 7 d, Fig. 8 d.
It is respectively compared Fig. 6 d and Fig. 6 b, Fig. 7 d and Fig. 7 b, Fig. 8 d and Fig. 8 b, wherein Fig. 6 (b), 7
(b), 8 (b) is the result images of the Dim targets detection obtained only with max-medium filter method.By
Figure is visible, and the present invention preferably inhibits texture while retaining well and strengthening Weak target signal
Complicated, large area consecutive variations and the cloudy background risen and fallen by force, can be accurately positioned and extract and flood
Infrared small object in complex background.
Although the present invention is disclosed as above with preferred embodiment, it is not for limiting the present invention, appointing
What those skilled in the art without departing from the spirit and scope of the present invention, may be by the disclosure above
Methods and techniques content makes possible variation and modification to technical solution of the present invention, therefore, every not take off
From the content of technical solution of the present invention, it is any that the technical spirit according to the present invention is made to above example
Simple modification, equivalent variation and modification, belong to the protection domain of technical solution of the present invention.
To sum up, advantages of the present invention includes：
First, the present invention carries out multiple dimensioned and multi-direction decomposition using the multiple anisotropic filter of tower antithesis tree,
Because it has the characteristics such as abundant yardstick, directional information and translation invariance so that infrared image is through place
Echo signal enhancing, contour edge information weaken after reason, improve the positioning precision of target.
Second, the present invention handles subband office on the basis of multiple dimensioned multi-direction decomposition using bilateral filtering
The interior coefficient value between neighborhood of portion's neighborhood, it is contemplated that the pixel of single isolated positions and the mutual pass of adjacent pixel
System and effect, take full advantage of the information such as regional area luminosity, gray scale and geometry similitude, very well
Ground retains and strengthens Weak target signal, is effectively improved the overall signal to noise ratio of image and contrast.
3rd, the present invention seeks the performance of optimum solution using artificial ant algorithm, in image preprocessing
On the basis of, the Weak target after effectively background is suppressed in image is extracted, and precise marking goes out target
Position in the picture.
Claims (5)
1. a kind of infrared image detection method of small target based on multiple dimensioned bilateral optimization, its feature exists
In, including：
(1) infrared image is read in；
(2) background suppression is carried out to the infrared image of reading, the background process of inhibition includes：
A. exploded view picture
The scale parameter and direction number of exploded view picture are set, using tower antithesis tree anisotropic filter group to reading in
Image is decomposed by setting scale parameter and direction initialization number is decomposed, and the picture size sum after decomposition at different levels is upper
The size of first order image, the number sum of the image after decomposition at different levels is the number of upper level image；
The tower antithesis tree anisotropic filter group is by translatable tower wave filter and an antithesis tree 2n passage
Anisotropic filter group realize；
B. the high frequency direction sub-band coefficients of all directions are changed using bilateral filtering method；
C. low frequency sub-band coefficient is changed using median filter method；
D. reconstructed image
Using the inverse transformation of tower antithesis tree anisotropic filter group step by step to the reconstruct of sub-band coefficients travel direction and chi
Degree reconstruct, obtains the background image of estimation；
E. target residual image is obtained；
The background image subtraction for the estimation that the infrared image of reading is obtained with reconstruct, is carried on the back
The result images that scape suppresses；
(3) Weak target is extracted
A. ant group algorithm is used, the result images after suppressing by the use of background calculate ant as heuristic information
State transition probability；
B. after being circulated throughout every time, pheromones are updated according to the Pheromone update mechanism in ant system；
Wherein, when kth ant is transferred to (i, j) place,Other situations are then
The update mechanism of traditional ant group algorithm pheromones can cause the pheromones of the former destination path accessed
It is intended to zero with the increase of cycle-index, all cycle-indexes is traveled through for this and will be believed after each circulation
The codomain scope of breath element is limited in [τmin, τmax] interval interior, then there is second of renewal of pheromones：
(4) result images are obtained
According to pheromones distribution segmentation image zooming-out target.
2. the infrared image Weak target inspection according to claim 1 based on multiple dimensioned bilateral optimization
Survey method, it is characterised in that the background histamine result is and right as the heuristic information in ant group algorithm
Pheromones carry out secondary update.
3. the infrared image Weak target inspection according to claim 1 based on multiple dimensioned bilateral optimization
Survey method, it is characterised in that the infrared image includes the Weak target for being submerged in strong varying background clutter,
The cloud that the strong varying background includes cloud layer edge, large area continuously rises and falls.
4. the infrared image Weak target inspection according to claim 1 based on multiple dimensioned bilateral optimization
Survey method, it is characterised in that the step of reading in infrared image includes：Gather infrared using thermal infrared imager
Image, computer is read in using computer keyboard and mouse by the infrared image.
5. the infrared image Weak target inspection according to claim 1 based on multiple dimensioned bilateral optimization
Survey method, it is characterised in that extracting mesh calibration method includes：The path of Ant Search is with cycle-index
Increase gradually restrains to target area, and it shows as the pheromones left on destination path in ant group algorithm
Significantly greater than other regions, after given number of iterations and cycle-index is reached, if the letter on node (i, j)
Breath element is met
τ″I, j≥τ′max
Then node (i, j) is labeled as target, is otherwise background.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110016175.8A CN106508046B (en) | 2011-12-28 | 2011-12-28 | One kind is based on multiple dimensioned bilateral optimized detection method of small target |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110016175.8A CN106508046B (en) | 2011-12-28 | 2011-12-28 | One kind is based on multiple dimensioned bilateral optimized detection method of small target |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106508046B true CN106508046B (en) | 2014-11-05 |
Family
ID=58359872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110016175.8A Active CN106508046B (en) | 2011-12-28 | 2011-12-28 | One kind is based on multiple dimensioned bilateral optimized detection method of small target |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106508046B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107403134A (en) * | 2017-05-27 | 2017-11-28 | 西安电子科技大学 | The multiple dimensioned method for detecting infrared puniness target in figure domain based on the side of partial gradient three |
| CN109767439A (en) * | 2019-01-10 | 2019-05-17 | 中国科学院上海技术物理研究所 | A kind of multiple dimensioned difference of self-adapting window and the object detection method of bilateral filtering |
| CN110782405A (en) * | 2019-10-14 | 2020-02-11 | 中国科学院光电技术研究所 | Point target and dark spot image background balancing method based on gradient identification |
| CN113536912A (en) * | 2021-06-09 | 2021-10-22 | 中国铁塔股份有限公司黑龙江省分公司 | Twin comparison same-class tower type early warning algorithm based on standard model |
| CN117315376A (en) * | 2023-11-28 | 2023-12-29 | 聊城莱柯智能机器人有限公司 | Machine learning-based mechanical part industrial quality inspection method |
-
2011
- 2011-12-28 CN CN201110016175.8A patent/CN106508046B/en active Active
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107403134A (en) * | 2017-05-27 | 2017-11-28 | 西安电子科技大学 | The multiple dimensioned method for detecting infrared puniness target in figure domain based on the side of partial gradient three |
| CN107403134B (en) * | 2017-05-27 | 2022-03-11 | 西安电子科技大学 | Local gradient trilateral-based image domain multi-scale infrared dim target detection method |
| CN109767439A (en) * | 2019-01-10 | 2019-05-17 | 中国科学院上海技术物理研究所 | A kind of multiple dimensioned difference of self-adapting window and the object detection method of bilateral filtering |
| CN109767439B (en) * | 2019-01-10 | 2023-05-05 | 中国科学院上海技术物理研究所 | Target detection method for multi-scale difference and bilateral filtering of self-adaptive window |
| CN110782405A (en) * | 2019-10-14 | 2020-02-11 | 中国科学院光电技术研究所 | Point target and dark spot image background balancing method based on gradient identification |
| CN110782405B (en) * | 2019-10-14 | 2022-10-18 | 中国科学院光电技术研究所 | Point target and dark spot image background balancing method based on gradient identification |
| CN113536912A (en) * | 2021-06-09 | 2021-10-22 | 中国铁塔股份有限公司黑龙江省分公司 | Twin comparison same-class tower type early warning algorithm based on standard model |
| CN117315376A (en) * | 2023-11-28 | 2023-12-29 | 聊城莱柯智能机器人有限公司 | Machine learning-based mechanical part industrial quality inspection method |
| CN117315376B (en) * | 2023-11-28 | 2024-02-13 | 聊城莱柯智能机器人有限公司 | Machine learning-based mechanical part industrial quality inspection method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yang et al. | Dense people counting using IR-UWB radar with a hybrid feature extraction method | |
| CN103473764B (en) | A kind of remote sensing image object variations detection method | |
| CN106508046B (en) | One kind is based on multiple dimensioned bilateral optimized detection method of small target | |
| Temlioglu et al. | A novel convolutional autoencoder-based clutter removal method for buried threat detection in ground-penetrating radar | |
| CN107563355A (en) | Hyperspectral abnormity detection method based on generation confrontation network | |
| CN102831598B (en) | Remote sensing image change detecting method with combination of multi-resolution NMF (non-negative matrix factorization) and Treelet | |
| CN107247930A (en) | SAR image object detection method based on CNN and Selective Attention Mechanism | |
| CN103839073B (en) | Polarization SAR image classification method based on polarization features and affinity propagation clustering | |
| CN103824088B (en) | SAR target variant recognition method based on multi-information joint dynamic sparse representation | |
| CN104200471B (en) | SAR image change detection based on adaptive weight image co-registration | |
| CN104835178A (en) | Low SNR(Signal to Noise Ratio) motion small target tracking and identification method | |
| CN102254319A (en) | Method for carrying out change detection on multi-level segmented remote sensing image | |
| CN107403134B (en) | Local gradient trilateral-based image domain multi-scale infrared dim target detection method | |
| CN103837889A (en) | Seismic attribute multi-source information fusion method | |
| CN102930275B (en) | Based on the characteristics of remote sensing image system of selection of Cramer ' s V index | |
| CN106886760A (en) | A kind of EO-1 hyperion Ship Detection combined based on empty spectrum information | |
| CN111667019A (en) | Hyperspectral image classification method based on deformable separation convolution | |
| CN104036461A (en) | Infrared complicated background inhibiting method based on combined filtering | |
| Li et al. | Infrared small target detection via center-surround gray difference measure with local image block analysis | |
| Yao et al. | Hyperspectral anomaly detection based on the bilateral filter | |
| Myint et al. | An evaluation of four different wavelet decomposition procedures for spatial feature discrimination in urban areas | |
| CN102609721A (en) | Remote sensing image clustering method | |
| CN112784777A (en) | Unsupervised hyperspectral image change detection method based on antagonistic learning | |
| Venkateswaran et al. | Performance analysis of k-means clustering for remotely sensed images | |
| Yang et al. | Radar target recognition using contourlet packet transform and neural network approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR03 | Grant of secret patent right | ||
| GRSP | Grant of secret patent right | ||
| DC01 | Secret patent status has been lifted | ||
| DCSP | Declassification of secret patent |