CN103761731A - Small infrared aerial target detection method based on non-downsampling contourlet transformation - Google Patents
Small infrared aerial target detection method based on non-downsampling contourlet transformation Download PDFInfo
- Publication number
- CN103761731A CN103761731A CN201410000396.XA CN201410000396A CN103761731A CN 103761731 A CN103761731 A CN 103761731A CN 201410000396 A CN201410000396 A CN 201410000396A CN 103761731 A CN103761731 A CN 103761731A
- Authority
- CN
- China
- Prior art keywords
- target
- pipeline
- image
- coefficient
- noise
- 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
Images
Abstract
The invention provides a small infrared aerial target detection method based on non-downsampling contourlet transformation. The method includes the following steps of 1, non-downsampling contourlet transformation, wherein non-downsampling contourlet transformation first-level decomposition is performed on a small infrared target image, and a band pass sub-band is discomposed into four-direction high-frequency sub-bands; 2, background suppression, wherein low-frequency influences are removed, and thresholding processing is performed on a high-frequency portion; 3, coefficient mapping, wherein coefficients left by the four-direction high-frequency sub-bands are mapped to a gray level space in a linear mode; 4, high-frequency image segmentation, wherein four-direction high-frequency sub-band images are segmented into binaryzation images; 5, binary high-frequency image noise reduction, wherein small bright noise points in the binary high-frequency images are eliminated; 6, detection of related small targets in dimension, wherein the four-direction high-frequency sub-band images get along with each other to obtain a small target single-frame detection result; 7, small target sequence detection, wherein comprehensive vote is performed on multi-frame images to intercept and capture small targets. According to the method, the problem of small aerial target detection under the complicated infrared background is solved.
Description
Technical field
The present invention relates to a kind of small IR target detection, particularly the aerial small target detecting method under a kind of Infrared Complex Background condition.
Background technology
For real application systems, how to give full play to the advantage of infrared target detection technique, strive for having become to determine in the relevant information that obtain to attack target the strongest opportunity the key factor of modern war victory or defeat.So, the problem that the detection that improves as far as possible target is all paid special attention to apart from Cheng Liao various countries researchist.Distance is far away, and the imaging area of target in detection means is less, and target will increase by the possibility of clutter and background influence, thereby detection difficulty also can strengthen.At remote tracking phase, target imaging in imaging system is the spot that acnode or several pixel form, and what in visual field, exist is chronic, and signal intensity is weak and easily by clutter, flooded.Battlefield surroundings generally comprises sea, heaven and earth face.In real battlefield surroundings, due to smog, cloud layer, the interface on the mountain range on earth's surface, texture, large building and air-sea ground all can make background complicated.These complicated backgrounds all can produce greatly and disturb the identification of moving target.In the infrared image generating in infrared remote aerial reconnaissance, the brightness of little target is usually above background luminance, and target area is little, and brightness changes also less.Sky background mainly consists of cloud layer, rain, mist etc., and noise is mainly the interference of infrared system inside.Infrared small target image background confusion area is large, and signal to noise ratio (S/N ratio) is lower, and its this spatial character cannot be applied traditional image processing detection technique of utilizing target sizes, shape and feature.How in Infrared Complex Background environment, accurately identifying the small and weak target of aerial signal becomes a difficult problem urgently to be resolved hurrily, and the solution of this problem is for increasing operational distance and increasing the reaction time, and the survival probability that improves one's own side has great importance.
Summary of the invention
Object of the present invention is just to provide a kind of infrared aerial small target detecting method based on non-down sampling contourlet transform, overcomes the impact of Infrared Complex Background and noise, solves the little target detection problems of infrared aerial under complex background condition.
The present invention is that the technical scheme that the deficiency that solves the problems of the technologies described above adopts is:
A kind of infrared aerial small target detecting method based on non-down sampling contourlet transform:
Step 1, non-down sampling contourlet transform: the little target image of infrared aerial is carried out to the decomposition of non-down sampling contourlet one-level, is low frequency part and HFS by picture breakdown, and wherein high-frequency band pass sub-band division is four direction;
Step 2, background suppress: remove the impact of low frequency part, noise and the background of calculating high-frequency sub-band suppress threshold value, and HFS coefficient is carried out to thresholding processing;
Step 3, coefficient mapping: the coefficient that high frequency is retained takes absolute value, linear mapping, to gray space, obtains the high frequency imaging of four direction;
Step 4, high frequency imaging are cut apart: utilize threshold method to cut apart high frequency imaging, four direction high-frequency sub-band images is divided into binary image;
Step 5, two-value high frequency imaging noise reduction: utilize the some bright spot noises in mathematical morphology open operator filtering binaryzation high frequency imaging;
The little target detection of correlativity in step 6, yardstick: four direction high frequency imaging phase with obtain little target single frame detection result;
Step 7, little target sequence detect: utilize the correlativity of little target position between consecutive frame, by multiple image, comprehensively decide by vote, realize the little target acquisition of infrared aerial.
Described low frequency part is background information, and HFS is little target information, noise and a small amount of background information.
Described HFS coefficient carries out thresholding disposal route, and the HFS obtaining in step 1 is carried out to wavelet transformation, and the coefficient after wavelet transformation is divided into two classes: first kind wavelet coefficient is only by obtaining after noise transformation, and amplitude is little and number is more; Equations of The Second Kind wavelet coefficient is obtained by little target information and a small amount of background information signal conversion, and amplitude is large and number is less.In little target image, the coefficient of echo signal is in Equations of The Second Kind coefficient.Like this, by the difference in wavelet coefficient amplitude, can construct a kind of noise-reduction method.A threshold value is set, and the wavelet coefficient that is greater than this threshold value belongs to Equations of The Second Kind coefficient, can retain; And the wavelet coefficient that is less than this threshold value is exactly first kind coefficient, can remove.The threshold expression that Donoho proposes is:
, wherein, N presentation video pixel count. σ represents that Noise Factor Standard is poor.This threshold method extends to non-down sampling contourlet transform noise reduction.
Described step 7 is to utilize the correlativity of target position between consecutive frame, the image that comprises final goal with acquisition by the comprehensive voting of multiple image.Pipeline filter multi frame detection method is according to the continuity of target travel, a space pipeline of setting up centered by suspected target on the locus of sequence image, calculate subsequently in pipeline window, belong to suspected target number of pixels and with given threshold, thereby judge a two field picture in pipeline, whether comprise target and determine position.
The specific implementation step that described little target sequence detects is,
Step 1, according to the duct length of setting, set up pipeline.For each two field picture that is about to enter pipeline, all to carry out phase and computing with the former frame image of processing through morphological dilations being present in pipeline, with further filtering noise.After pipeline is set up, similar and pipeline organization, often newly enters a two field picture, and the image that completes detection just flows out pipeline, and remaining image flows successively.
After step 2, pipeline are set up, for the single frames bianry image that enters at first pipeline, sign 8 connected regions wherein, using this as candidate target region.
Step 3, for each candidate target region, calculate its barycenter, using it as pipeline center, in whole pipeline, calculate the number of suspicious object pixel by (5) formula:
(5)
Wherein, l is duct length, and m is pipe diameter, (x
0, y
0) be candidate target region center-of-mass coordinate, i is frame number in pipeline, M is suspicious object pixel count.For further filtering random noise and fixed position noise, determine that the condition of target is: M is greater than a certain thresholding and same position place candidate target number is less than a certain thresholding.As condition meets, record object position.
Step 4, renewal pipeline, identify current 8 connected regions that enter the earliest piping drawing picture, as candidate target region, forwards step 3 to, until all images are disposed.
Beneficial effect of the present invention is: first the present invention utilizes non-down sampling contourlet transform to carry out one-level decomposition to the little target image of infrared aerial, is removing low frequency part impact and is considering on the basis of high-frequency sub-band oriented energy Background suppression filtering noise; Then, utilize the correlativity of little target in the yardstick of non-down sampling contourlet territory, the single frames of realizing little target just detects; The 3rd, the correlativity based on moving small target at adjacent interframe movement, decides by vote the accurate intercepting and capturing of further rejecting false-alarm and realizing infrared small target by multiframe.The present invention can overcome the impact of Infrared Complex Background and noise, has effectively solved the little target detection problems of the infrared aerial under complex background condition.
Accompanying drawing explanation
Fig. 1 is that the 40th frame of infrared video 1 of the present invention is through the result of correlation detection in the inhibition of non-down sampling contourlet transform background, Threshold segmentation, yardstick;
Fig. 2 is that the 40th frame of infrared video 1 of the present invention is through the result of correlation detection in the inhibition of wavelet transformation background, Threshold segmentation, yardstick;
Fig. 3 is that the 60th frame of infrared video 1 of the present invention is through the result of correlation detection in the inhibition of non-down sampling contourlet transform background, Threshold segmentation, yardstick;
Fig. 4 is that the 60th frame of infrared video 1 of the present invention is through the result of correlation detection in the inhibition of wavelet transformation background, Threshold segmentation, yardstick;
Fig. 5 is that the 20th frame of infrared video 2 of the present invention is through the result of correlation detection in the inhibition of non-down sampling contourlet transform background, Threshold segmentation, yardstick;
Fig. 6 is that the 20th frame of infrared video 2 of the present invention is through the result of correlation detection in the inhibition of wavelet transformation background, Threshold segmentation, yardstick;
Fig. 7 is the 40th frame process conventional algorithm background inhibition of infrared video 1;
Fig. 8 is the 60th frame process conventional algorithm background inhibition of infrared video 1;
Fig. 9 is the 20th frame process conventional algorithm background inhibition of infrared video 2.
Embodiment
Basic ideas of the present invention: the little target detection of infrared aerial is the isolated singularity in judgement image, and wavelet transformation can be distinguished the target (singular point) that is positioned at HFS and the background that is positioned at low frequency part well, thereby be widely used aspect little target detection.The existing small target detecting method based on small echo is all, at frequency domain, image wavelet coefficient is carried out to respective handling, reaches reconstruct again after the object of Background suppression, then adopts other method further to process.But the conversion of two-dimentional separable wavelets only has level, vertical, three, diagonal angle direction, the defect in directivity and anisotropy makes the wavelet transformation directional information in presentation video well.
Profile wave convert, a kind of multiple dimensioned, two dimensional image analysis method local, directivity.Profile wave convert has not only been inherited the multi-resolution time-frequency analysis feature of wavelet transformation, and have a good anisotropic character, can carry out better rarefaction representation to image, can hold exactly image geometry structural information, effectively capture the profile in natural image.Profile wave convert becomes the logical directional subband of band on each yardstick picture breakdown.Laplacian pyramid (Laplacian pyramid, LP) turriform is decomposed the multiple dimensioned decomposition that has been used to profile wave convert.Every one-level LP decomposes and will produce the low pass sampling of a upper level signal and the band reduction of fractions to a common denominator amount being obtained by the difference of low pass sampling and upper level signal.The multiple dimensioned decomposition of next stage is that iteration is carried out in the low pass sampling producing.The spectrum division wedgewise frequency subband that anisotropic filter group (directional filter bank, DFB) decomposes LP the logical image of band obtaining, completes the Directional Decomposition on each yardstick of profile ripple.But due to the existence of down-sampling process, profile wave convert lacks translation invariance, image is processed Hou edge can produce pseudo-Gibbs distortion.
Propose subsequently, non-down sampling contourlet transform, it had both had advantages of profile wave convert, had again translation invariance, thereby can effectively solve pseudo-this problem of Gibbs distortion.Non-down sampling contourlet transform is the same with profile wave convert, is also to adopt the two iterative filter group structures that consist of LP conversion and DFB.That both differences are non-down sampling contourlet transform employing is non-sampling LP and non-sampling DFB, during conversion, by the tower-like wave filter of non-sampling, by picture breakdown, be first low frequency part and HFS, then by non-sampling side tropism's bank of filters, HFS be decomposed into several directions.Non-down sampling contourlet transform removed that LP decomposes and DFB decomposition in the signal up-sampling (interpolation) before filtered down-sampling (extraction) and integrated filter by analysis, and change into, corresponding wave filter is carried out to up-sampling, then signal is carried out to analysis filtered and integrated filter.Non-down sampling contourlet transform can not only distinguish each frequency band of image, and multidirectional and translation invariance strengthens its details protective capability, therefore, if applied it in the middle of the little target detection of infrared aerial, should access the result that is better than wavelet transformation.The non-sample direction wave filter of non-sampling turriform wave filter and four direction carries out convolution with former infrared image and the logical component image of band respectively, generates the high fdrequency component of low frequency component and four direction.Non-down sampling contourlet transform low frequency component has mainly reflected the feature of background, and comprises little target information, noise and a small amount of background information (having comprised the marginal information that there is no filtering) in detected image in each direction of high frequency.Because background information has certain style characteristic, the position distribution correlativity in the different directions high fdrequency component after decomposition a little less than; And little target does not have style characteristic, the position distribution correlativity in the different directions high fdrequency component after non-down sampling contourlet transform decomposes is stronger.By high frequency all directions testing result is associated, can further eliminates background information and strengthen target.Based on above analysis, first the present invention utilizes the multiple dimensioned decomposition of non-down sampling contourlet transform to the little target image of infrared aerial, then utilize the correlation properties of little target factor in yardstick to realize single frame detection, last binding sequence imagery exploitation pipeline filter completes the Detection task of the little target of infrared aerial under complex background.
Detailed process of the present invention
:
Infrared background suppresses and noise reduction: the coefficient after wavelet transformation is divided into two classes: first kind wavelet coefficient is only by obtaining after noise transformation, and amplitude is little and number is more; Equations of The Second Kind wavelet coefficient is obtained by signal conversion, and amplitude is large and number is less.In little target image, the coefficient of echo signal is in Equations of The Second Kind coefficient.Like this, by the difference in wavelet coefficient amplitude, can construct a kind of noise-reduction method.A threshold value is set, and the wavelet coefficient that is greater than this threshold value belongs to Equations of The Second Kind coefficient, can retain; And the wavelet coefficient that is less than this threshold value is exactly first kind coefficient, can remove.The threshold expression that Donoho proposes is:
Wherein, N presentation video pixel count. σ represents that Noise Factor Standard is poor.This threshold method extends to non-down sampling contourlet transform noise reduction.
Detector in infrared imaging guidance system is the main source of system noise, is the principal element that affects infrared image quality, and the noise that its intensity generally produces much larger than other link is also difficult to most overcome simultaneously.The noise of detector itself is unavoidable, according to the mechanism of its generation, can be divided into shot noise, thermonoise, photon noise, generation recombination noise and 1/f noise etc.The detector current output that wherein shot noise, thermonoise, photon noise and generation recombination noise produce is a stochastic process, by the approximate Gaussian distributed of central limit theorem.1/f noise is a kind of current noise of infrared eye low frequency part, and as its name suggests, 1/f noise and frequency are inversely proportional to, and when frequency is during higher than certain frequency, compares and can ignore with other noise.Therefore, can think and affect the noise of infrared image quality Gaussian distributed.
Non-down sampling contourlet transform has decorrelation character, and this has guaranteed that the concentration of energy of image after conversion is on limited coefficient in transform domain, and the amplitude of all the other most of coefficient in transform domain is close to zero; White Gaussian noise is still white noise after conversion, and energy distribution is on all coefficient in transform domain.Noise after non-down sampling contourlet transform
kthe energy of all directions of layer can be thought approximately equalised, so the energy of more noisy image conversion coefficient in all directions is also just equivalent to relatively expect the energy of image conversion coefficient in all directions.When decomposition scale is determined, the coefficient gross energy under each yardstick has just been determined.The proportion that coefficient energy in certain direction accounts for whole Scale energy is larger, and the profile details of key diagram picture in this direction is many, and what to the threshold value in this direction, should arrange is lower, can retain many profile details like this; In like manner, the energy comparison in certain direction is little, has also just illustrated that the profile details of image in this direction is fewer, and what to the threshold value in this direction, should arrange is higher, like this can reasonable denoising.For this reason, can consider oriented energy factor pair
trevise:
Infrared background inhibition and noise reduction concrete steps are as follows:
1. pairs of small target infrared images of Step carry out the decomposition of non-down sampling contourlet transform one-level, by picture breakdown, are low frequency part and four direction HFS.
Step 2. removes the impact of low frequency part, suppresses most of background information, by (2) formula calculating noise and background, suppresses threshold value, and high frequency coefficient is partly carried out to thresholding processing by (3) formula:
Step 3. takes absolute value left coefficient, and is mapped to gray space by (4) formula, obtains the high frequency imaging of four direction:
High frequency imaging is cut apart: for little target image, utilize imaging sensor sensitivity, resolution, can image-forming range and the possible information such as actual size of target, the size of estimating target in image, i.e. shared number of pixels m
0.The high frequency imaging gray scale intermediate value of take is starting point, and on histogram, first statistical number of forward lookup is m
0gray-scale value as segmentation threshold.Utilize threshold method to cut apart high frequency imaging, four direction high-frequency sub-band images is divided into binary image.The high frequency imaging of binaryzation comprises some bright spot noises, and morphology opening operation can filtering be less than the bright spot of structural element, therefore utilize morphology to open the further filtering noise of operator.The structural element size of modified opening operator is less than target area.
Infrared small target single frame detection: utilize little target factor in the correlativity of high frequency direction, the high frequency imaging of four direction is carried out to phase and computing, form Preliminary detection result.
Infrared small target Sequence Detection: utilize the correlativity of target position between consecutive frame, the image that comprises final goal with acquisition by the comprehensive voting of multiple image.Pipeline filter multi frame detection method is according to the continuity of target travel, a space pipeline of setting up centered by suspected target on the locus of sequence image, calculate subsequently in pipeline window, belong to suspected target number of pixels and with given threshold, thereby judge a two field picture in pipeline, whether comprise target and determine position.Concrete steps are as follows:
Step 1. sets up pipeline according to the duct length of setting.For each two field picture that is about to enter pipeline, all to carry out phase and computing with the former frame image of processing through morphological dilations being present in pipeline, with further filtering noise.After pipeline is set up, similar and pipeline organization, often newly enters a two field picture, and the image that completes detection just flows out pipeline, and remaining image flows successively.
After Step 2. pipelines are set up, for the single frames bianry image that enters at first pipeline, sign 8 connected regions wherein, using this as candidate target region.
Step 3., for each candidate target region, calculates its barycenter, usings it as pipeline center, by (5) formula, calculates the number of suspicious object pixel in whole pipeline:
(5)
Wherein, l is duct length, and m is pipe diameter, (x
0, y
0) be candidate target region center-of-mass coordinate, i is frame number in pipeline, M is suspicious object pixel count.For further filtering random noise and fixed position noise, determine that the condition of target is: M is greater than a certain thresholding and same position place candidate target number is less than a certain thresholding.As condition meets, record object position.
Step 4. upgrades pipeline, identifies current 8 connected regions that enter the earliest piping drawing picture, as candidate target region, forwards step 3 to, until all images are disposed.
In 44 FLIR video sequences that experiment has adopted U.S.Army Aviation and Missile Command (AMCOM) to provide, intercept 200 groups of image sequences and detect, every group of 100 frames.Wherein, the result of the 20th frame process non-down sampling contourlet transform of the 40th frame, 60 frames and the infrared video 2 of infrared video 1 and the inhibition of wavelet transformation background, Threshold segmentation, the interior correlation detection of yardstick as Figure 1-3.In non-down sampling contourlet transform, turriform wave filter is got " 9-7 ", anisotropic filter is got " and pkva ", carry out one-level decomposition, direction number is 4.The wavelet transformation base comparing is " haar ".
From Fig. 1-3, can find out, the effect that four high frequency direction backgrounds of non-down sampling contourlet transform suppress will obviously be better than three high frequency direction of wavelet transformation.After Threshold segmentation, morphologic filtering, four high frequency direction have all retained target information preferably and position is fixed, and residual noise point is less and position is random.Further utilize the correlativity in little target factor yardstick, all successfully realize the aerial little target detection of single frames.
As shown in Figure 4, in order to verify background inhibition of the present invention, three frame original images are processed through spatial domain high-pass filtering, morphology top cap conversion filtering, Butterworth high-pass filtering, wavelet thresholding method based on direction coefficient energy respectively, provided the signal to noise ratio (S/N ratio) of the little target image of three frames after each algorithm process as table 1.Therefrom can find out, spatial domain high-pass filtering background is residual at most, morphology top cap converter technique and Butterworth method target a little less than, in three two field pictures, the snr value of inventive method has all reached maximum.
The snr value of each algorithm of table 1
For infrared video 1 and 2, according to motion relevance between little target frame, further carry out Sequence Detection, result is as table 2.
The Sequence Detection result of each algorithm of table 2
Spatial domain high pass is because background suppresses residual more, so verification and measurement ratio is low and false alarm rate is high; Morphology top cap and Butterworth high pass after background suppresses, echo signal a little less than, verification and measurement ratio still can, false alarm rate is lower; Inventive method is all better than additive method in two indexs.
The present invention has studied the infrared aerial small target detecting method under complex background in non-down sampling contourlet territory.Characteristic of the present invention is, considering that non-down sampling contourlet decomposes on the basis of all directions coefficient energy, noise and background edge coefficient are suppressed, subsequently, not by inverse transformation reconstructed image, but be gray level image by all directions non-down sampling contourlet coefficient mapping, and then complete testing process by a series of spatial processings.The detection method conventional with some compared, and the present invention has some superiority aspect background inhibition, can obtain relatively high signal to noise ratio (S/N ratio).The little target image of actual infrared aerial detects explanation, and the present invention can obtain testing result accurately and effectively.In addition, checking by experiment, conventional little algorithm of target detection cannot detect target substantially in a two field picture, needs the accumulation of multiple image to detect.And the present invention can detect target by single-frame images under many circumstances, even if need multiple frame cumulation, the frame number needing also seldom, thereby also has certain advantage aspect detection speed.
In sum, the present invention is more practical than current other system, simple and have good background inhibition and detect performance.
Claims (5)
1. the infrared aerial small target detecting method based on non-down sampling contourlet transform:
Step 1, non-down sampling contourlet transform: the little target image of infrared aerial is carried out to the decomposition of non-down sampling contourlet one-level, is low frequency part and HFS by picture breakdown, and wherein high-frequency band pass sub-band division is four direction;
Step 2, background suppress: remove the impact of low frequency part, noise and the background of calculating high-frequency sub-band suppress threshold value, and HFS coefficient is carried out to thresholding processing;
Step 3, coefficient mapping: the coefficient that high frequency is retained takes absolute value, linear mapping, to gray space, obtains the high frequency imaging of four direction;
Step 4, high frequency imaging are cut apart: utilize threshold method to cut apart high frequency imaging, four direction high-frequency sub-band images is divided into binary image;
Step 5, two-value high frequency imaging noise reduction: utilize the some bright spot noises in mathematical morphology open operator filtering binaryzation high frequency imaging;
The little target detection of correlativity in step 6, yardstick: four direction high frequency imaging phase with obtain little target single frame detection result;
Step 7, little target sequence detect: utilize the correlativity of little target position between consecutive frame, by multiple image, comprehensively decide by vote, realize the little target acquisition of infrared aerial.
2. a kind of infrared aerial small target detecting method based on non-down sampling contourlet transform as claimed in claim 1, is characterized in that: described low frequency part is background information, and HFS is little target information, noise and a small amount of background information.
3. a kind of infrared aerial small target detecting method based on non-down sampling contourlet transform as claimed in claim 2, it is characterized in that: described HFS coefficient carries out thresholding disposal route and is, the HFS obtaining in step 1 is carried out to wavelet transformation, coefficient after wavelet transformation is divided into two classes: first kind wavelet coefficient is only by obtaining after noise transformation, and amplitude is little and number is more; Equations of The Second Kind wavelet coefficient is obtained by little target information and a small amount of background information signal conversion, amplitude is large and number is less, in little target image, the coefficient of echo signal is in Equations of The Second Kind coefficient, by the difference in wavelet coefficient amplitude, can construct a kind of noise-reduction method, a threshold value is set, and the wavelet coefficient that is greater than this threshold value belongs to Equations of The Second Kind coefficient, can retain; And the wavelet coefficient that is less than this threshold value is exactly first kind coefficient, can remove, threshold expression is:, wherein, N presentation video pixel count. σ represents that Noise Factor Standard is poor.
4. a kind of infrared aerial small target detecting method based on non-down sampling contourlet transform as claimed in claim 1, it is characterized in that: described step 7 is to utilize the correlativity of target position between consecutive frame, the image that comprises final goal with acquisition by the comprehensive voting of multiple image, pipeline filter multi frame detection method is according to the continuity of target travel, a space pipeline of setting up centered by suspected target on the locus of sequence image, calculate subsequently in pipeline window, belong to suspected target number of pixels and with given threshold, whether thereby judging a two field picture in pipeline comprises target and determines position.
5. a kind of infrared aerial small target detecting method based on non-down sampling contourlet transform as claimed in claim 4, is characterized in that: the specific implementation step that described little target sequence detects is,
Step 1, according to the duct length of setting, set up pipeline, for each two field picture that is about to enter pipeline, all to carry out phase and computing with the former frame image of processing through morphological dilations being present in pipeline, with further filtering noise, after pipeline is set up, similar and pipeline organization, often newly enters a two field picture, the image that completes detection just flows out pipeline, and remaining image flows successively;
After step 2, pipeline are set up, for the single frames bianry image that enters at first pipeline, sign 8 connected regions wherein, using this as candidate target region;
Step 3, for each candidate target region, calculate its barycenter, using it as pipeline center, in whole pipeline, calculate the number of suspicious object pixel by (5) formula:
Wherein, l is duct length, and m is pipe diameter, (x
0, y
0) be candidate target region center-of-mass coordinate, i is frame number in pipeline, M is suspicious object pixel count, for further filtering random noise and fixed position noise, the condition of determining target is: M is greater than a certain thresholding and same position place candidate target number is less than a certain thresholding, as condition meets, record object position;
Step 4, renewal pipeline, identify current 8 connected regions that enter the earliest piping drawing picture, as candidate target region, forwards step 3 to, until all images are disposed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410000396.XA CN103761731A (en) | 2014-01-02 | 2014-01-02 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410000396.XA CN103761731A (en) | 2014-01-02 | 2014-01-02 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103761731A true CN103761731A (en) | 2014-04-30 |
Family
ID=50528964
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410000396.XA Pending CN103761731A (en) | 2014-01-02 | 2014-01-02 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103761731A (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104036461A (en) * | 2014-06-10 | 2014-09-10 | 河南科技大学 | Infrared complicated background inhibiting method based on combined filtering |
CN104599273A (en) * | 2015-01-22 | 2015-05-06 | 南京理工大学 | Wavelet multi-scale crossover operation based sea-sky background infrared small target detection method |
CN104766289A (en) * | 2015-03-20 | 2015-07-08 | 华南理工大学 | Image denoising and compressing method |
CN104899866A (en) * | 2015-05-05 | 2015-09-09 | 河南三联网络技术有限公司 | Intelligent infrared small target detection method |
CN105528783A (en) * | 2015-12-01 | 2016-04-27 | 上海电机学院 | Multi-level compact texture feature extraction method and image texture segmentation method |
CN106019241A (en) * | 2016-05-13 | 2016-10-12 | 中国航空工业集团公司西安飞机设计研究所 | Single target intercept method for radar simulators |
CN103971371B (en) * | 2014-05-17 | 2017-02-22 | 北京航空航天大学 | spatial point target detection method based on focal plane modulation characteristics |
CN107563982A (en) * | 2017-09-12 | 2018-01-09 | 新疆大学 | A kind of Enhancement Methods about Satellite Images |
CN107742282A (en) * | 2017-11-13 | 2018-02-27 | 中国人民解放军国防科技大学 | Superimposed image preprocessing method based on main direction suppression high-pass filter |
CN108154507A (en) * | 2017-12-29 | 2018-06-12 | 长春师范大学 | Screwed pipe foreign matter detection system |
CN108198198A (en) * | 2017-12-22 | 2018-06-22 | 湖南源信光电科技股份有限公司 | Single frames infrared small target detection method based on wavelet transformation and Steerable filter |
CN108280460A (en) * | 2017-12-04 | 2018-07-13 | 西安电子科技大学 | Based on the SAR vehicle target recognition methods for improving convolutional neural networks |
CN108364277A (en) * | 2017-12-20 | 2018-08-03 | 南昌航空大学 | A kind of infrared small target detection method of two-hand infrared image fusion |
CN109409192A (en) * | 2018-08-27 | 2019-03-01 | 浙江大丰实业股份有限公司 | Stage laser navigation verifies platform immediately |
CN109523575A (en) * | 2018-11-12 | 2019-03-26 | 南通理工学院 | Method for detecting infrared puniness target |
CN109559286A (en) * | 2018-11-19 | 2019-04-02 | 电子科技大学 | A kind of variance gradient constraint method infrared image edge holding denoising method |
CN109657639A (en) * | 2018-12-29 | 2019-04-19 | 中国电子科技集团公司第二十七研究所 | A kind of Situation Awareness System and method based on panoramic vision |
CN111429479A (en) * | 2020-03-26 | 2020-07-17 | 中国科学院长春光学精密机械与物理研究所 | Space target identification method based on image integral mean value |
CN111951299A (en) * | 2020-07-01 | 2020-11-17 | 中国科学院上海技术物理研究所 | Infrared aerial target detection method |
CN112215893A (en) * | 2020-10-28 | 2021-01-12 | 安徽农业大学 | Method, device and equipment for determining target two-dimensional center coordinate point and ranging system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1136948A1 (en) * | 2000-03-21 | 2001-09-26 | European Community | Method of multitime filtering coherent-sensor detected images |
CN102324021A (en) * | 2011-09-05 | 2012-01-18 | 电子科技大学 | Infrared dim-small target detection method based on shear wave conversion |
CN103400383A (en) * | 2013-07-22 | 2013-11-20 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection |
-
2014
- 2014-01-02 CN CN201410000396.XA patent/CN103761731A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1136948A1 (en) * | 2000-03-21 | 2001-09-26 | European Community | Method of multitime filtering coherent-sensor detected images |
CN102324021A (en) * | 2011-09-05 | 2012-01-18 | 电子科技大学 | Infrared dim-small target detection method based on shear wave conversion |
CN103400383A (en) * | 2013-07-22 | 2013-11-20 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection |
Non-Patent Citations (4)
Title |
---|
刘兴淼 等: "基于非下采样Contourlet变换的红外小目标检测方法", 《红外》 * |
刘刚 等: "基于小波变换和管道滤波的红外空中小目标检测", 《计算机工程与应用》 * |
刘刚 等: "红外空中小目标小波域检测方法", 《火力与指挥控制》 * |
康志亮: "基于小波的红外图像增强算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103971371B (en) * | 2014-05-17 | 2017-02-22 | 北京航空航天大学 | spatial point target detection method based on focal plane modulation characteristics |
CN104036461B (en) * | 2014-06-10 | 2017-01-04 | 河南科技大学 | A kind of Infrared Complex Background suppressing method based on Federated filter |
CN104036461A (en) * | 2014-06-10 | 2014-09-10 | 河南科技大学 | Infrared complicated background inhibiting method based on combined filtering |
CN104599273A (en) * | 2015-01-22 | 2015-05-06 | 南京理工大学 | Wavelet multi-scale crossover operation based sea-sky background infrared small target detection method |
CN104599273B (en) * | 2015-01-22 | 2017-07-28 | 南京理工大学 | Sea and sky background infrared small target detection method based on multi-scale wavelet crossing operation |
CN104766289A (en) * | 2015-03-20 | 2015-07-08 | 华南理工大学 | Image denoising and compressing method |
CN104899866A (en) * | 2015-05-05 | 2015-09-09 | 河南三联网络技术有限公司 | Intelligent infrared small target detection method |
CN104899866B (en) * | 2015-05-05 | 2018-03-30 | 河南三联网络技术有限公司 | A kind of intelligentized infrared small target detection method |
CN105528783A (en) * | 2015-12-01 | 2016-04-27 | 上海电机学院 | Multi-level compact texture feature extraction method and image texture segmentation method |
CN106019241A (en) * | 2016-05-13 | 2016-10-12 | 中国航空工业集团公司西安飞机设计研究所 | Single target intercept method for radar simulators |
CN106019241B (en) * | 2016-05-13 | 2018-10-09 | 中国航空工业集团公司西安飞机设计研究所 | A kind of radar simulator single goal method for interception |
CN107563982A (en) * | 2017-09-12 | 2018-01-09 | 新疆大学 | A kind of Enhancement Methods about Satellite Images |
CN107742282A (en) * | 2017-11-13 | 2018-02-27 | 中国人民解放军国防科技大学 | Superimposed image preprocessing method based on main direction suppression high-pass filter |
CN108280460A (en) * | 2017-12-04 | 2018-07-13 | 西安电子科技大学 | Based on the SAR vehicle target recognition methods for improving convolutional neural networks |
CN108280460B (en) * | 2017-12-04 | 2021-07-27 | 西安电子科技大学 | SAR vehicle target identification method based on improved convolutional neural network |
CN108364277A (en) * | 2017-12-20 | 2018-08-03 | 南昌航空大学 | A kind of infrared small target detection method of two-hand infrared image fusion |
CN108198198A (en) * | 2017-12-22 | 2018-06-22 | 湖南源信光电科技股份有限公司 | Single frames infrared small target detection method based on wavelet transformation and Steerable filter |
CN108154507A (en) * | 2017-12-29 | 2018-06-12 | 长春师范大学 | Screwed pipe foreign matter detection system |
CN108154507B (en) * | 2017-12-29 | 2023-06-13 | 长春师范大学 | Foreign matter detection system for threaded pipe |
CN109409192A (en) * | 2018-08-27 | 2019-03-01 | 浙江大丰实业股份有限公司 | Stage laser navigation verifies platform immediately |
CN109523575A (en) * | 2018-11-12 | 2019-03-26 | 南通理工学院 | Method for detecting infrared puniness target |
CN109559286A (en) * | 2018-11-19 | 2019-04-02 | 电子科技大学 | A kind of variance gradient constraint method infrared image edge holding denoising method |
CN109559286B (en) * | 2018-11-19 | 2022-12-06 | 电子科技大学 | Variance gradient constraint method infrared image edge preserving denoising method |
CN109657639A (en) * | 2018-12-29 | 2019-04-19 | 中国电子科技集团公司第二十七研究所 | A kind of Situation Awareness System and method based on panoramic vision |
CN111429479B (en) * | 2020-03-26 | 2022-10-11 | 中国科学院长春光学精密机械与物理研究所 | Space target identification method based on image integral mean value |
CN111429479A (en) * | 2020-03-26 | 2020-07-17 | 中国科学院长春光学精密机械与物理研究所 | Space target identification method based on image integral mean value |
CN111951299B (en) * | 2020-07-01 | 2022-09-16 | 中国科学院上海技术物理研究所 | Infrared aerial target detection method |
CN111951299A (en) * | 2020-07-01 | 2020-11-17 | 中国科学院上海技术物理研究所 | Infrared aerial target detection method |
CN112215893A (en) * | 2020-10-28 | 2021-01-12 | 安徽农业大学 | Method, device and equipment for determining target two-dimensional center coordinate point and ranging system |
CN112215893B (en) * | 2020-10-28 | 2022-10-28 | 安徽农业大学 | Method, device and equipment for determining target two-dimensional center coordinate point and ranging system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103761731A (en) | Small infrared aerial target detection method based on non-downsampling contourlet transformation | |
Nasiri et al. | Infrared small target enhancement based on variance difference | |
CN109460764B (en) | Satellite video ship monitoring method combining brightness characteristics and improved interframe difference method | |
CN109919870B (en) | SAR image speckle suppression method based on BM3D | |
CN102609701B (en) | Remote sensing detection method based on optimal scale for high-resolution SAR (synthetic aperture radar) | |
CN111027496B (en) | Infrared dim target detection method based on space-time joint local contrast | |
CN107507209B (en) | Printogram extraction method of polarized SAR image | |
Yang et al. | Multiscale facet model for infrared small target detection | |
Li et al. | Road lane detection with gabor filters | |
CN106886747B (en) | It is a kind of based on extension wavelet transformation complex background under Ship Detection | |
Lipschutz et al. | New methods for horizon line detection in infrared and visible sea images | |
CN110400294B (en) | Infrared target detection system and detection method | |
CN104036461B (en) | A kind of Infrared Complex Background suppressing method based on Federated filter | |
Wang et al. | A robust infrared dim target detection method based on template filtering and saliency extraction | |
CN110660065B (en) | Infrared fault detection and identification algorithm | |
CN107220628B (en) | Method for detecting infrared interference source | |
CN111666854A (en) | High-resolution SAR image vehicle target detection method fusing statistical significance | |
Li et al. | A small target detection algorithm in infrared image by combining multi-response fusion and local contrast enhancement | |
CN105184804A (en) | Sea surface small target detection method based on airborne infrared camera aerially-photographed image | |
CN105469428B (en) | A kind of detection method of small target based on morphologic filtering and SVD | |
Zhao et al. | An adaptation of CNN for small target detection in the infrared | |
CN102722879A (en) | SAR (synthetic aperture radar) image despeckle method based on target extraction and three-dimensional block matching denoising | |
CN109002777A (en) | A kind of infrared small target detection method towards complex scene | |
CN106778822B (en) | Image straight line detection method based on funnel transformation | |
CN109461171A (en) | The small IR targets detection algorithm of DoG filtering is improved based on multichannel |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20140430 |