CN109886218A - SAR image Ship Target Detection method based on super-pixel statistics diversity - Google Patents
SAR image Ship Target Detection method based on super-pixel statistics diversity Download PDFInfo
- Publication number
- CN109886218A CN109886218A CN201910142161.7A CN201910142161A CN109886218A CN 109886218 A CN109886218 A CN 109886218A CN 201910142161 A CN201910142161 A CN 201910142161A CN 109886218 A CN109886218 A CN 109886218A
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- value
- image
- tcr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention belongs to Radar Technology fields, disclose a kind of SAR image Ship Target Detection method based on super-pixel statistics diversity.This method comprises: obtaining SAR image to be detected, super-pixel segmentation is carried out to SAR image and obtains W super-pixel；Secondly based on Gamma distribution it is assumed that calculate the corresponding Gamma distribution of each super-pixel form parameter and inverse scale parameter；The global contrast angle value drawn game portion contrast value for calculating each super-pixel, further calculates to obtain the corresponding TCR enhancement value of the super-pixel, using TCR enhancement value as the intensity value of each pixel in the super-pixel, to obtain TCR enhancing image；Finally, calculating detection threshold value T, binaryzation is carried out to TCR image, obtains the corresponding bianry image of TCR enhancing image.The present invention can preferably improve TCR based on super-pixel statistics diversity, so as to realize higher target detection rate with lower false alarm rate, promote target detection performance.
Description
Technical field
The present invention relates to Radar Technology fields, more particularly to the SAR image Ship Target based on super-pixel statistics diversity
Detection method.
Background technique
Synthetic aperture radar (full name in English: Synthetic Aperture Radar, english abbreviation: SAR) system not by
The limitation of the conditions such as illumination, weather can carry out round-the-clock, round-the-clock observation to interested target, obtain high-resolution
Two-dimensional SAR image is a kind of important earth observation means, is all widely applied in civil and military field.
It is grinding for field of radar based on the target detection of SAR image as the key technology in automatic Target Recognition System
Study carefully hot spot.Traditional constant false alarm rate (full name in English: Constant False-Alarm Rate, english abbreviation: CFAR) method,
As unit average constant false alarm (full name in English: Cell-Averaging Constant False-Alarm Rate, english abbreviation:
CA-CFAR) method, using the grey value difference of object pixel and clutter pixel in SAR gray level image, set suitable threshold value into
Row target detection.With the development of SAR technical level, the resolution ratio of SAR image is continuously improved, although this help to obtain more
Fine target information, but difficulty also is brought to CFAR method simultaneously.Since target size is long-range in High Resolution SAR Images
In distance by radar resolution cell, multiple scattering centers of target expand to different distance unit, and backward energy is dispersed, and cause mesh
The gray value for marking pixel rises and falls, and leads to target-clutter contrast (full name in English: Target to Clutter Rate, English contracting
Write: TCR) it reduces, the lower object pixel of the gray value when being detected using tradition CFAR method in target area can quilt
Missing inspection, to cause occur targeted fractured and discontinuous phenomenon in testing result.
In addition, the target in High Resolution SAR image shows complicated shape and structure relative to low resolution SAR image
Feature, it is not only the simple set of high luminance pixels point, and the spatial relationship between the pixel of target more embodies the structure of target
And shape feature, and existing CFAR method does not consider the sky between pixel mostly using pixel as the basic unit of target detection
Between relationship.In order to solve this problem, the concept of super-pixel is proposed, super-pixel is according to the similarity degree between pixel to figure
Topography's block that pixel as in is grouped.In general, the target in High Resolution SAR image can be divided into one or
Multiple adjacent super-pixel, using super-pixel as the basic unit of target detection, can not only obtain mesh relative to single pixel
More useful statistics and structural information are marked, and since the number of super-pixel is much smaller than the sum of all pixels in image, is conducive to
Reduce the operand of target detection process.
To realize SAR image offshore Ship Target Detection, there is article to propose a kind of super-pixel notable figure method, this method benefit
With pixel mean intensity in super-pixel, i.e. square of gray value, global and local significant characteristics are generated.This method is intended to mention
High TCR carries out target detection on this basis, and rejects false-alarm using contextual information.However, this method depends on super-pixel
The mean intensity of middle pixel can go out in object detection results when background super-pixel has similar intensity with target super-pixel
Existing false-alarm and missing inspection, target detection performance decline.
Summary of the invention
The embodiment of the present invention provides a kind of SAR image Ship Target Detection method based on super-pixel statistics diversity,
TCR can be preferably improved based on super-pixel statistics diversity, so as to realize higher target inspection with lower false alarm rate
Survey rate promotes target detection performance.
In order to achieve the above objectives, the embodiment of the present invention adopts the following technical scheme that
Step 1 obtains SAR image to be detected, carries out super-pixel segmentation to SAR image to be detected, obtains W super pictures
Element；
Step 2, based on Gamma distribution it is assumed that calculate the corresponding Gamma distribution of each super-pixel form parameter and inverse scale
Parameter；
Step 3, using the different value of statistics between each super-pixel and other W-1 super-pixel, calculate each super-pixel
Global contrast angle value；Wherein, the different value of statistics between any two super-pixel is by each super-pixel pair in two super-pixel
The form parameter for the Gamma distribution answered and inverse scale parameter are calculated；Using each super-pixel and its all neighbouring super pixels it
Between the different value of statistics, calculate the local contrast angle value of each super-pixel；For any one super-pixel, according to the super-pixel
Global contrast angle value drawn game portion contrast value is calculated corresponding TCR enhancement value, and then surpasses picture using TCR enhancement value as this
The intensity value of each pixel in element, to obtain TCR enhancing image；
Step 4 determines detection threshold value T；For any pixel point in TCR enhancing image, if the intensity value of the pixel
More than or equal to detection threshold value T, then the gray scale of the pixel is set as 255；If the intensity value of the pixel is less than detection threshold value T,
The gray scale of the pixel is then set as 0；The all pixels point in TCR enhancing image is traversed to get corresponding to TCR enhancing image
Bianry image, the corresponding region of pixel that wherein gray value is 255 are Ship Target, and the pixel that gray value is 0 is corresponding
Region is image background.
SAR image Ship Target Detection method provided by the invention based on super-pixel statistics diversity, first to be checked
It surveys SAR image and carries out super-pixel segmentation；It is then based on gamma distributional assumption, calculates the shape of the corresponding Gamma distribution of each super-pixel
Shape parameter and inverse scale parameter；Further joined using the form parameter and inverse scale of the corresponding gamma distribution of each super-pixel
Number, calculates the global contrast angle value drawn game portion contrast value of each super-pixel；According to the global contrast angle value of each super-pixel and
Local contrast angle value obtains the TCR enhancement value of each super-pixel, using the TCR enhancement value of each super-pixel as every in super-pixel
The intensity value of a pixel obtains TCR enhancing image；Calculate the corresponding form parameter of TCR enhancing image and inverse scale parameter meter
Calculation obtains detection threshold value；Binaryzation finally is carried out to TCR enhancing image using detection threshold value, obtains the corresponding two-value of TCR image
Image, the corresponding region of pixel that wherein gray value is 255 are Ship Target, the corresponding region of pixel that gray value is 0
For image background.Method provided by the invention is different based on the statistics calculated between every two super-pixel based on gamma distributional assumption
Value, and TCR enhancing figure is finally obtained based on super-pixel statistics diversity, TCR can be preferably improved, is further able to lower
False alarm rate realize higher target detection rate, promote target detection performance.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the SAR image Ship Target Detection method provided in an embodiment of the present invention that diversity is counted based on super-pixel
Flow diagram；
Fig. 2 is that the TCR obtained using different equivalent depending on the emulation SAR image of number enhances image comparison；Wherein Fig. 2 (a)-
(d) it is target pixel points and the corresponding gamma distribution of background clutter pixel, Fig. 2 (b) that corresponding equivalent number, which is 1, Fig. 2 (a),
For the emulation SAR image generated using the gamma distribution of target and background in Fig. 2 (a), the super-pixel point that Fig. 2 (c) is Fig. 2 (b)
It cuts as a result, Fig. 2 (d) is that the corresponding TCR of Fig. 2 (b) enhances image；Wherein the corresponding equivalent number of Fig. 2 (e)-(h) is 4, Fig. 2 (e)
For the corresponding gamma distribution of target and background clutter, Fig. 2 (f) is to be generated using the gamma distribution of target and background in Fig. 2 (e)
SAR image is emulated, Fig. 2 (g) is the super-pixel segmentation of Fig. 2 (f) as a result, Fig. 2 (h) is the corresponding TCR enhancing image of Fig. 2 (f)；
Fig. 3 is to utilize the method for the present invention, CA-CFAR method and super-pixel notable figure method to India's harbour SAR image
Comparison diagram；Wherein, Fig. 3 (a) is the SAR image at India harbour, and Fig. 3 (b) is ground truth figure, and Fig. 3 (c) is to utilize the present invention
The global super-pixel TCR enhancing figure that the method that embodiment provides obtains, Fig. 3 (d) is to utilize side provided in an embodiment of the present invention
The local super-pixel TCR enhancing figure that method obtains, the TCR enhancing figure that Fig. 3 (e) is obtained using method provided in an embodiment of the present invention
Picture, Fig. 3 (f) are to obtain the corresponding bianry image of Fig. 3 (a) using method provided in an embodiment of the present invention；Fig. 3 (g) is super-pixel
The TCR of notable figure method enhances image, and Fig. 3 (h) is the corresponding bianry image of Fig. 3 (a) obtained using super-pixel notable figure method；Figure
3 (i) be the corresponding bianry image of Fig. 3 (a) obtained using CA-CFAR method；
Fig. 4 is to utilize the method for the present invention, CA-CFAR method and super-pixel notable figure method to Singapore's harbour SAR image
Comparison diagram；Wherein, Fig. 4 (a) is the SAR image at Singapore harbour, and Fig. 4 (b) is ground truth figure, and Fig. 4 (c) is to utilize this
The obtained global super-pixel TCR enhancing figure of method that inventive embodiments provide, Fig. 4 (d) is to provide using the embodiment of the present invention
The obtained local super-pixel TCR enhancing figure of method, Fig. 4 (e) enhanced using the TCR that method provided in an embodiment of the present invention obtains
Image, Fig. 4 (f) are to obtain the corresponding bianry image of Fig. 4 (a) using method provided in an embodiment of the present invention；Fig. 4 (g) is super picture
The TCR of plain notable figure method enhances image, and Fig. 4 (h) is the corresponding bianry image of Fig. 4 (a) obtained using super-pixel notable figure method；
Fig. 4 (i) is the corresponding bianry image of Fig. 4 (a) obtained using CA-CFAR method.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the SAR image Ship Target Detection method provided in an embodiment of the present invention that diversity is counted based on super-pixel
Flow diagram.
Referring to Fig. 1, the SAR image Ship Target Detection side provided in an embodiment of the present invention based on super-pixel statistics diversity
Method specifically includes the following steps:
Step 1 obtains SAR image to be detected, carries out super-pixel segmentation to SAR image to be detected, obtains W super pictures
Element.
Further, step 1 specifically includes:
Step 1.1 enables I indicate the number of iterations, initializes I=1；SAR image to be detected is obtained, SAR to be detected is schemed
Rectangular block R as being divided into N number of S × S1, R2..., Rn..., RN, take rectangular block R1, R2..., Rn..., RNGeometric center
c1, c2..., cn..., cNAs the cluster centre of SAR image, and with label 1,2 ..., n ..., N label；
Wherein,For super-pixel size, H is the line number of pixel in SAR image, and L is pixel in SAR image
Columns；
Step 1.2, the label for calculating ith pixel point in SAR image, specifically: in obtaining with ith pixel point and being
The heart, the square neighborhood that 2S is side length, by K cluster centre c in square neighborhoodi1,ci2,…,cik,…,ciKAs i-th
The candidate cluster center of a pixel calculates separately the difference value of ith pixel point Yu each candidate cluster center, by difference value
Label of the corresponding label in the smallest candidate cluster center as ith pixel point；Whole pixels in SAR image are traversed,
Obtain the label of each pixel in SAR image；
Wherein, i=1,2 ..., H × L；
Preferably, the difference value for calculating ith pixel point and each candidate cluster center, specifically includes:
Obtain the image block u centered on ith pixel pointiWith with candidate cluster center cikCentered on image block uik,
Image block uiWith image block uikIt is the image block of 5 × 5 pixels；
Calculate image block uiWith image block uikStrength difference valueAnd image block uiWith image block uikBetween
Space length
Further, image block u is calculatediWith image block uikStrength difference valueIt specifically includes:
According to preset formula, image block u is calculatediWith image block uikStrength difference valueWherein, in advance
If formula are as follows:
Wherein,ximFor
Image block uiIn m-th of pixel intensity, xikmFor image block uikIn m-th of pixel intensity,For image block uiWith
Image block uikPixel and concentrate j-th of pixel intensity, M be image block uiThe sum of middle pixel, image block uik
Sum and the image block u of middle pixeliMiddle pixel it is total equal, L=α be SAR image equivalent number.
It should be noted that the formula for calculating the strength difference value of any two image block is according to the picture in SAR image
What the hypothesis that the intensity of vegetarian refreshments obeys gamma distribution obtained, wherein parameterMaximal possibility estimation by seeking each image block
The mean value of middle pixel intensity obtains.
Preferably, image block u is calculatediWith image block uikBetween space length d (i, cik), it specifically includes:
According to formulaImage block u is calculatediWith image block uikIt
Between space length d (i, cik)；
Wherein, (ai,bi) it is coordinate of the pixel i in SAR image,For candidate cluster center cikIn SAR
Coordinate in image.
Using strength difference value and space length, image block u is calculatediWith image block uikDifference valueBy image block uiWith image block uikDifference value as ith pixel point and
Candidate cluster center cikDifference value, wherein λ=2S is the weight that accounts in difference value of adjustment space distance；
All candidate cluster centers for traversing ith pixel point obtain ith pixel point and each candidate cluster center
Difference value.
Wherein, i=1,2 ..., H × L；
Step 1.3, the set for being constituted the identical pixel of label in SAR image obtain P as a super-pixel
Super-pixel；For each super-pixel, wherein row coordinate of each pixel in SAR image and column coordinate are obtained, will be owned
The row coordinate of cluster centre of the mean value of the corresponding row coordinate of pixel as the super-pixel sits the corresponding column of all pixels point
Column coordinate of the target mean value as the cluster centre of the super-pixel, to obtain the cluster centre of the super-pixel；By P super-pixel
Cluster centre of the cluster centre as SAR image；It enables I add 1, judges whether I is less than or equal to 10, if it is go to step
1.2, if it is not, then going to step 1.4；
Step 1.4 obtains W super-pixel Q1, Q2..., Qw..., QW, wherein W is the super-pixel obtained after 10 iteration
Number.
It should be noted that the rectangular block number N of finally obtained super-pixel number W and initial setting not necessarily phase
Together, the reason is that being iterated to calculate by 10 times, may occur without corresponding pixel in some labels.
Step 2, based on Gamma distribution it is assumed that calculate the corresponding Gamma distribution of each super-pixel form parameter and inverse scale
Parameter.
Further, in step 2, any super-pixel Q is calculatedwThe form parameter of corresponding Gamma distribution and inverse scale are joined
Number, specifically includes:
Calculate super-pixel QwCorresponding single order sample logarithmic momentWith second order sample logarithmic moment
Wherein, NwIndicate super-pixel QwThe sum of middle pixel, xwn'Indicate super-pixel QwIn n-th ' a pixel intensity
Value, n'=1,2 ..., Nw,w∈[1,2,…,W]；
Calculate super-pixel QwCorresponding single order logarithm cumulative amountWith second order logarithm cumulant
Establish equation groupSolve system of equation obtains super-pixel QwCorresponding Gamma distribution
Form parameterWith inverse scale parameterWherein, Ψ () is polynary polygamma function, and Ψ (1) represents Ψ
The first derivative of () function.
It should be noted that the foundation based on gamma distributional assumption is the priori statistical model based on SAR image herein,
It can be concluded that the background clutter pixel of SAR image and target pixel points obey gamma distribution.
Step 3, using the different value of statistics between each super-pixel and other W-1 super-pixel, calculate each super-pixel
Global contrast angle value；Wherein, the different value of statistics between any two super-pixel is by each super-pixel pair in two super-pixel
The form parameter for the Gamma distribution answered and inverse scale parameter are calculated；Using each super-pixel and its all neighbouring super pixels it
Between the different value of statistics, calculate the local contrast angle value of each super-pixel；For any one super-pixel, according to the super-pixel
Global contrast angle value drawn game portion contrast value is calculated corresponding TCR enhancement value, and then surpasses picture using TCR enhancement value as this
The intensity value of each pixel in element, to obtain TCR enhancing image；
Further, any super-pixel Q is calculatedwGlobal contrast angle value, local contrast angle value and corresponding TCR enhancing
Value, specifically includes:
According to formulaSuper-pixel Q is calculatedwGlobal contrast
Angle value Sglobal(Qw)；
Wherein, w ∈ [1,2 ..., W], r ∈ [1,2 ..., W] and r ≠ w；D(pQw,pQr) it is super-pixel QwWith super-pixel QrIt
Between statistics diversity value；
According to formulaSuper-pixel Q is calculatedwPart
Contrast value Slocal(Qw)；
Wherein, D (pQw,pQs) it is super-pixel Qw and super-pixel QsBetween the different value of statistics,Indicate super-pixel Qw and super-pixel QsBetween space length power
Weight, Dspatial(Qw,Qs) indicate super-pixel QwWith super-pixel QsBetween Euclidean distance,(aw,bw) it is super-pixel QwGeometric center in SAR image
Coordinate, (as,bs) it is super-pixel QsCoordinate of the geometric center in SAR image, Ω (Qw) indicate super-pixel QwAdjacent super picture
The set of element,S ∈ [1,2 ..., W] and s ≠ w；
According to formula Ssp(Qw)=Sglobal(Qw)·Slocal(Qw) super-pixel Q is calculatedwCorresponding TCR enhancement value Ssp
(Qw)。
It should be noted that using the global contrast angle value of each super-pixel as the intensity of each pixel in the super-pixel
Value, traverses the global super-pixel TCR enhancing figure of the available SAR image of all super-pixel；The part of each super-pixel is right
Intensity value than angle value as each pixel in the super-pixel, the part for traversing the available SAR image of all super-pixel are super
Pixel TCR enhancing figure.
Preferably, any super-pixel Q is calculatedwWith super-pixel QrBetween statistics diversity value, specifically include:
According to formulaSuper-pixel Q is calculatedwWith
Super-pixel QrBetween statistics diversity value D (pQw, pQr)；
Wherein, w ∈ [1,2 ..., W], r ∈ [1,2 ..., W] and r ≠ w；
Γ () is the gamma function of standard,For super-pixel QwThe form parameter of corresponding gamma distribution and inverse scale parameter,For super-pixel QrThe form parameter of corresponding gamma distribution and inverse scale parameter.
Step 4 calculates detection threshold value T, for any pixel point in TCR enhancing image, if the intensity value of the pixel
More than or equal to detection threshold value T, then the gray scale of the pixel is set as 255.If the intensity value of the pixel is less than detection threshold value T,
The gray scale of the pixel is then set as 0.The all pixels point in TCR enhancing image is traversed to get corresponding to TCR enhancing image
Bianry image, the corresponding region of pixel that wherein gray value is 255 are Ship Target, and the pixel that gray value is 0 is corresponding
Region is image background.
Further, it in step 4, determines detection threshold value T, specifically includes:
Calculate 1 rank sample logarithmic moment of TCR enhancing imageWith 2 rank sample logarithmic moments
Wherein, xn″It is the n-th " intensity value of a pixel, n "=1,2 ..., N in TCR enhancing imageTCR, NTCRFor TCR enhancing
The sum of pixel in image；
Calculate 1 rank logarithm cumulant of TCR enhancing imageWith 2 rank logarithm cumulants
Establish equation groupSolve system of equation obtains the corresponding gamma distribution of TCR enhancing image
Form parameterWith inverse scale parameter
Wherein, Ψ () is polynary polygamma function, and Ψ (1) represents the first derivative of Ψ () function；
Based on constant false alarm rate principle and Gamma distribution it is assumed that design false alarm rate Pfa, utilize the corresponding gamma of TCR enhancing image
The form parameter of distribution and inverse scale parameter establish false alarm rate PfaWith the equation of detection threshold value T
It solves the equation and obtains detection threshold value T；
Wherein, Γ () is the gamma function of standard.
SAR image Ship Target Detection method provided by the invention based on super-pixel statistics diversity, first to be checked
It surveys SAR image and carries out super-pixel segmentation；It is then based on gamma distributional assumption, calculates the shape of the corresponding Gamma distribution of each super-pixel
Shape parameter and inverse scale parameter；Further joined using the form parameter and inverse scale of the corresponding gamma distribution of each super-pixel
Number, calculates the global contrast angle value drawn game portion contrast value of each super-pixel；According to the global contrast angle value of each super-pixel and
Local contrast angle value obtains the TCR enhancement value of each super-pixel, using the TCR enhancement value of each super-pixel as every in super-pixel
The intensity value of a pixel obtains TCR enhancing image；Calculate the corresponding form parameter of TCR enhancing image and inverse scale parameter meter
Calculation obtains detection threshold value；Binaryzation finally is carried out to TCR enhancing image using detection threshold value, obtains the corresponding two-value of TCR image
Image, the corresponding region of pixel that wherein gray value is 255 are Ship Target, the corresponding region of pixel that gray value is 0
For image background.Method provided by the invention is different based on the statistics calculated between every two super-pixel based on gamma distributional assumption
Value, and TCR enhancing figure is finally obtained based on super-pixel statistics diversity, TCR can be preferably improved, is further able to lower
False alarm rate realize higher target detection rate, promote target detection performance.
Further, the above-mentioned beneficial effect of the present invention is verified below by way of emulation experiment.
Emulation experiment operation platform: Matlab R2014a, Intel (R) Core (TM) i7-4790CPU@3.6GHz, memory
8GB。
Experiment one:
Fig. 2 (a) is that background clutter pixel obeys form parameter α=L=1 and the gamma of inverse scale parameter β=0.5 is distributed
The gamma for obeying form parameter α=L=1 and inverse scale parameter β=0.2 with target pixel points is distributed.Fig. 2 (e) is background clutter
Pixel obeys the gamma distribution of form parameter α=L=4 and inverse scale parameter β=0.5 and target pixel points obey form parameter
The gamma of α=L=4 and inverse scale parameter β=0.2 is distributed.Fig. 2 (b) and Fig. 2 (f) is respectively that (a) and Fig. 2 (e) are raw according to fig. 2
At emulation SAR image, emulate SAR image size be 200 × 200 pixels, be arranged super-pixel size S=20, utilize this hair
The method that bright embodiment provides is respectively processed to obtain Fig. 2 (c), Fig. 2 (d), Fig. 2 (g) and Fig. 2 to Fig. 2 (b) and Fig. 2 (f)
(h), wherein Fig. 2 (c) is the super-pixel segmentation of Fig. 2 (b) as a result, Fig. 2 (d) is the corresponding TCR enhancing image of Fig. 2 (b), Fig. 2
(g) super-pixel segmentation for being Fig. 2 (f) is as a result, Fig. 2 (h) is the corresponding TCR enhancing image of Fig. 2 (f).TCR value is introduced as TCR
The judgment criteria of reinforcing effect, table 1 are to utilize side provided in an embodiment of the present invention when equivalent number L takes 1,2,3 and 4 respectively
The TCR value for the TCR enhancing image that method obtains:
Table 1
Method provided in an embodiment of the present invention can significantly improve TCR as can be seen from Table 1, and when equivalent number L is got over
When big, TCR value is bigger, shows that TCR reinforcing effect is better.
Experiment two:
Fig. 3 (a) is that the SAR image at India harbour stops two in the image scene near harbour as shown in Fig. 3 (a)
A Ship Target, shown in ground truth such as Fig. 3 (b).Method provided in an embodiment of the present invention, super-pixel notable figure side is respectively adopted
Method and CA-CFAR method carry out target detection, and in method provided in an embodiment of the present invention, super-pixel size S=85 is arranged, point
Fig. 3 (c)-Fig. 3 (f) is not obtained, wherein Fig. 3 (a) is the SAR image at India harbour, and Fig. 3 (b) is ground truth figure, Fig. 3
It (c) is the global super-pixel TCR enhancing figure obtained using method provided in an embodiment of the present invention, Fig. 3 (d) is to utilize the present invention
The local super-pixel TCR enhancing figure that the method that embodiment provides obtains, Fig. 3 (e) are obtained using method provided in an embodiment of the present invention
The TCR enhancing image arrived, Fig. 3 (f) is to obtain the corresponding bianry image of Fig. 3 (a) using method provided in an embodiment of the present invention；Figure
3 (g) enhance image for the TCR of super-pixel notable figure method, and Fig. 3 (h) is that the Fig. 3 (a) obtained using super-pixel notable figure method is corresponded to
Bianry image；Fig. 3 (i) is the corresponding bianry image of Fig. 3 (a) obtained using CA-CFAR method.
Comparison diagram 3 (e) and Fig. 3 (g), it can be seen that it is almost invisible in the target subject of the middle circles mark of Fig. 3 (g),
Cause target area in Fig. 3 (h) to lack, it can be deduced that the targets improvement effect of method provided in an embodiment of the present invention compared with
It is good, i.e., it is showed in enhancing target and in terms of inhibiting background clutter more preferable.Comparison diagram 3 (f) and Fig. 3 (i), it can be seen that the present invention
The method that embodiment provides can completely detect Ship Target, and not have false-alarm in testing result, target in Fig. 3 (i)
Region includes cavity and crack, and the false-alarm pixel largely marked with rectangle frame occurs in land area.
Experiment three:
Fig. 4 (a) is that the SAR image at Singapore harbour is stopped in the image scene near harbour as shown in Fig. 4 (a)
It is multiple that there is various sizes of Ship Target, and there is the large area land area with strong backscatter intensity, ground truth
As shown in Fig. 4 (b).Method, the significant drawing method of super-pixel and CA-CFAR method provided in an embodiment of the present invention is respectively adopted to carry out
Target detection is arranged super-pixel size S=25, respectively obtains Fig. 4 (c)-Fig. 4 in method provided in an embodiment of the present invention
(f), wherein Fig. 4 (a) is the SAR image at India harbour, and Fig. 4 (b) is ground truth figure, and Fig. 4 (c) is real using the present invention
The global super-pixel TCR enhancing figure that the method for applying example offer obtains, Fig. 4 (d) is to utilize method provided in an embodiment of the present invention
Obtained local super-pixel TCR enhancing figure, Fig. 4 (e) enhance image using the TCR that method provided in an embodiment of the present invention obtains,
Fig. 4 (f) is to obtain the corresponding bianry image of Fig. 4 (a) using method provided in an embodiment of the present invention；Fig. 4 (g) is that super-pixel is significant
The TCR of figure method enhances image, and Fig. 4 (h) is the corresponding bianry image of Fig. 4 (a) obtained using super-pixel notable figure method；Fig. 4 (i)
For the corresponding bianry image of Fig. 4 (a) obtained using CA-CFAR method.
Comparison diagram 4 (e) and Fig. 4 (g), it can be seen that although super-pixel notable figure method can preferably inhibit the land of large area
Ground region, however the intensity of some Ship Targets is also inhibited, and shows as the naval vessel gone out in Fig. 4 (h) with circles mark
There is missing inspection in target.Although comparison diagram 4 (f) and Fig. 4 (i) can be seen that CA-CFAR method is capable of detecting when all targets,
It is that there are some false-alarms generated by land area for target area in Fig. 4 (i).
In order to quantitatively assess target detection performance, the actually detected rate DR of target and practical false alarm rate FAR are defined respectively:
The actually detected rate DR=N of targetdt/NT, NdtFor the object pixel points being correctly detecting, NTIt is total for object pixel
Number；Practical false alarm rate FAR=Ndc/NC, NdcFor the clutter sum of all pixels for being mistaken for object pixel in clutter pixel, NCFor clutter
Sum of all pixels.Method provided in an embodiment of the present invention is respectively adopted, super-pixel notable figure method and CA-CFAR are to Fig. 3 (a) and Fig. 4
(a) the actually detected rate DR of the corresponding target of the image handled and practical false alarm rate FAR are as shown in table 2:
Table 2
| Method | Method provided in an embodiment of the present invention | Super-pixel notable figure method | CA-CFAR |
| Index | DR/FAR (%) | DR/FAR (%) | DR/FAR (%) |
| Fig. 3 (a) | 91.04/0.28 | 47.01/0.18 | 40.19/1.56 |
| Fig. 4 (a) | 95.60/1.23 | 71.19/1.90 | 85.93/1.40 |
From Table 2, it can be seen that the method for the present invention can obtain highest verification and measurement ratio, while keeping lower false alarm rate.
For Fig. 3 (a), CA-CFAR method used time 17.2s, and method provided in an embodiment of the present invention uses MATLAB and C
++ hybrid programming, super-pixel segmentation used time 1.5s, image enhancement and detection used time 1.2s greatly improve the effect of target detection
Rate can satisfy the requirement of real-time.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light
The various media that can store program code such as disk.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (9)
1. a kind of SAR image Ship Target Detection method based on super-pixel statistics diversity, which is characterized in that the method packet
Include following steps:
Step 1 obtains SAR image to be detected, carries out super-pixel segmentation to SAR image to be detected, obtains W super-pixel；
Step 2, based on Gamma distribution it is assumed that calculating the form parameter of the corresponding Gamma distribution of each super-pixel and inverse scale is joined
Number；
Step 3, using the different value of statistics between each super-pixel and other W-1 super-pixel, calculate the complete of each super-pixel
Office's contrast value；Wherein, the different value of statistics between any two super-pixel is by each super-pixel pair in described two super-pixel
The form parameter for the Gamma distribution answered and inverse scale parameter are calculated；Using each super-pixel and its all neighbouring super pixels it
Between the different value of statistics, calculate the local contrast angle value of each super-pixel；For any one super-pixel, according to the super-pixel
Corresponding TCR enhancement value is calculated in global contrast angle value drawn game portion contrast value, and then using the TCR enhancement value as this
The intensity value of each pixel in super-pixel, to obtain TCR enhancing image；
Step 4 determines detection threshold value T；For any pixel point in TCR enhancing image, if the intensity value of the pixel
More than or equal to the detection threshold value T, then the gray scale of the pixel is set as 255；If the intensity value of the pixel is less than the inspection
Threshold value T is surveyed, then the gray scale of the pixel is set as 0；The all pixels point in the TCR enhancing image is traversed to get described in
TCR enhances the corresponding bianry image of image, and the corresponding region of pixel that wherein gray value is 255 is Ship Target, gray value
It is image background for the 0 corresponding region of pixel.
2. the method according to claim 1, wherein the step 1 specifically includes:
Step 1.1 enables I indicate the number of iterations, initializes I=1；SAR image to be detected is obtained, the SAR to be detected is schemed
Rectangular block R as being divided into N number of S × S1, R2..., Rn..., RN, take the rectangular block R1, R2..., Rn..., RNGeometry
Center c1, c2..., cn..., cNIt is marked as the cluster centre of the SAR image, and with label 1,2 ..., n ..., N；
Wherein,For super-pixel size, H is the line number of pixel in SAR image, and L is the columns of pixel in SAR image；
Step 1.2, the label for calculating ith pixel point in the SAR image, specifically: in obtaining with ith pixel point and being
The heart, the square neighborhood that 2S is side length, by K cluster centre c in the square neighborhoodi1, ci2..., cik..., ciKAs
The candidate cluster center of ith pixel point calculates separately the difference value of ith pixel point Yu each candidate cluster center, by institute
State label of the corresponding label in the smallest candidate cluster center of difference value as ith pixel point；It traverses in the SAR image
Whole pixels obtain the label of each pixel in SAR image；
Wherein, i=1,2 ..., H × L；
Step 1.3, using the identical pixel of label is constituted in the SAR image set as a super-pixel, obtain P
Super-pixel；For each super-pixel, wherein row coordinate of each pixel in the SAR image and column coordinate are obtained, it will
The row coordinate of cluster centre of the mean value of the corresponding row coordinate of all pixels point as the super-pixel, all pixels point is corresponding
Column coordinate of the mean value of column coordinate as the cluster centre of the super-pixel, to obtain the cluster centre of the super-pixel；It is super by P
Cluster centre of the cluster centre of pixel as the SAR image；It enables I add 1, judges whether I is less than or equal to 10, if it is
Step 1.2 is gone to, if it is not, then going to step 1.4；
Step 1.4 obtains W super-pixel Q1, Q2..., QW, wherein W is the number of the super-pixel obtained after 10 iteration.
3. according to the method described in claim 2, it is characterized in that, in the step 1.2, the calculating ith pixel point
With the difference value at each candidate cluster center, specifically include:
Obtain the image block u centered on ith pixel pointiWith with candidate cluster center cikCentered on image block uik, described
Image block uiWith described image block uikIt is the image block of 5 × 5 pixels；
Calculate described image block uiWith described image block uikStrength difference valueAnd described image block uiWith it is described
Image block uikBetween space length d (i, cik)；
Using the strength difference value and space length, described image block u is calculatediWith described image block uikDifference valueBy the described image block uiWith described image block uikDifference value as institute
State ith pixel point and candidate cluster center cikDifference value, wherein λ=2S is that adjustment space distance accounts in difference value
Weight；
All candidate cluster centers for traversing the ith pixel point obtain in the ith pixel point and each candidate cluster
The difference value of the heart.
4. according to the method described in claim 3, it is characterized in that, the calculating described image block uiWith described image block uik's
Strength difference valueIt specifically includes:
According to preset formula, described image block u is calculatediWith described image block uikStrength difference valueWherein,
The preset formula are as follows:
Wherein,ximFor image block
uiIn m-th of pixel intensity, xikmFor image block uikIn m-th of pixel intensity,For image block uiAnd image block
uikPixel and concentrate j-th of pixel intensity, M be image block uiThe sum of middle pixel, image block uikMiddle pixel
Sum and the image block u of pointiMiddle pixel it is total equal, L=α be the SAR image equivalent number.
5. according to the method described in claim 3, it is characterized in that, the calculating image block uiWith described image block uikBetween
Space length d (i, cik), it specifically includes:
According to formulaDescribed image block u is calculatediWith described image block
uikBetween space length d (i, cik)；
Wherein, (ai, bi) it is coordinate of the pixel i in SAR image,For candidate cluster center cikIn SAR image
Coordinate.
6. the method according to claim 1, wherein calculating any super-pixel QwThe shape of corresponding Gamma distribution
Parameter and inverse scale parameter, specifically include:
Calculate super-pixel QwCorresponding single order sample logarithmic momentWith second order sample logarithmic moment
Wherein, NwIndicate super-pixel QwThe sum of middle pixel, xwn′Indicate super-pixel QwIn the n-th ' a pixel intensity value, n '
=1,2 ..., Nw, w ∈ [1,2 ..., W]；
Calculate super-pixel QwCorresponding single order logarithm cumulative amountWith second order logarithm cumulant
Establish equation groupSolve system of equation obtains super-pixel QwThe shape of corresponding Gamma distribution
ParameterWith inverse scale parameterWherein, Ψ () is polynary polygamma function, and Ψ (1) represents Ψ () function
First derivative.
7. the method according to claim 1, wherein calculating any super-pixel QwGlobal contrast angle value, part it is right
Than angle value and corresponding TCR enhancement value, specifically include:
According to formulaSuper-pixel Q is calculatedwGlobal contrast angle value
Sglobal(Qw)；
Wherein, w ∈ [1,2 ..., W], r ∈ [1,2 ..., W] and r ≠ w；For super-pixel QwWith super-pixel QrIt
Between statistics diversity value；
According to formulaSuper-pixel Q is calculatedwLocal contrast
Value Slocal(Qw)；
Wherein,For super-pixel QwWith super-pixel QsBetween the different value of statistics,Indicate super-pixel QwWith super-pixel QsBetween space length power
Weight, Dspatial(Qw, Qs) indicate super-pixel QwWith super-pixel QsBetween Euclidean distance,(aw, bw) it is super-pixel QwGeometric center in SAR image
Coordinate, (αs, bs) it is super-pixel QsCoordinate of the geometric center in SAR image, Ω (Qw) indicate super-pixel QwAdjacent super picture
The set of element,S ∈ [1,2 ..., W] and s ≠ w；
According to formula Ssp(Qw)=Sglobal(Qw)·Slocal(Qw) super-pixel Q is calculatedwCorresponding TCR enhancement value Ssp(Qw)。
8. according to the method described in claim 1, calculating any super-pixel QwWith super-pixel QrBetween statistics diversity value, tool
Body includes:
According to formulaSuper-pixel Q is calculatedwWith super picture
Plain QrBetween statistics diversity value
Wherein, w ∈ [1,2 ..., W], r ∈ [1,2 ..., W] and r ≠ w；
Γ () is the gamma function of standard,For super-pixel QwCorresponding gal
The form parameter of horse distribution and inverse scale parameter,For super-pixel QrThe form parameter of corresponding gamma distribution and inverse scale
Parameter.
9. the method according to claim 1, wherein the determining detection threshold value T is specific to wrap in the step 4
It includes:
Calculate 1 rank sample logarithmic moment of the TCR enhancing imageWith 2 rank sample logarithmic moments
Wherein, xn″It is the n-th " intensity value of a pixel, n "=1,2 ..., N in the TCR enhancing imageTcR, NTCRFor the TCR
Enhance the sum of pixel in image；
Calculate 1 rank logarithm cumulant of the TCR enhancing imageWith 2 rank logarithm cumulants
It is thin to establish equationSolve system of equation obtains the shape of the corresponding gamma distribution of the TCR enhancing image
Shape parameterWith inverse scale parameter
Wherein, Ψ () is polynary polygamma function, and Ψ (1) represents the first derivative of Ψ () function；
Based on constant false alarm rate principle and Gamma distribution it is assumed that design false alarm rate Pfa, utilize the corresponding gamma of TCR enhancing image
The form parameter of distribution and inverse scale parameter establish the false alarm rate PfaWith the equation of the detection threshold value TIt solves the equation and obtains detection threshold value T；
Wherein, Γ () is the gamma function of standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910142161.7A CN109886218B (en) | 2019-02-26 | 2019-02-26 | SAR image ship target detection method based on superpixel statistic dissimilarity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910142161.7A CN109886218B (en) | 2019-02-26 | 2019-02-26 | SAR image ship target detection method based on superpixel statistic dissimilarity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886218A true CN109886218A (en) | 2019-06-14 |
| CN109886218B CN109886218B (en) | 2022-12-23 |
Family
ID=66929467
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910142161.7A Active CN109886218B (en) | 2019-02-26 | 2019-02-26 | SAR image ship target detection method based on superpixel statistic dissimilarity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886218B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112164087A (en) * | 2020-10-13 | 2021-01-01 | 北京无线电测量研究所 | Super-pixel segmentation method and device based on edge constraint and segmentation boundary search |
| CN112163450A (en) * | 2020-08-24 | 2021-01-01 | 中国海洋大学 | Based on S3High-frequency ground wave radar ship target detection method based on D learning algorithm |
| CN112183518A (en) * | 2020-09-25 | 2021-01-05 | 伏羲九针智能科技(北京)有限公司 | Vein target point automatic determination method, device and equipment |
| CN112528468A (en) * | 2020-11-20 | 2021-03-19 | 南京航空航天大学 | Label antenna reverse design method based on electromagnetic field SAR image processing |
| CN112766287A (en) * | 2021-02-05 | 2021-05-07 | 清华大学 | SAR image ship target detection acceleration method based on density examination |
| CN113362293A (en) * | 2021-05-27 | 2021-09-07 | 西安理工大学 | SAR image ship target rapid detection method based on significance |
| CN113379694A (en) * | 2021-06-01 | 2021-09-10 | 大连海事大学 | Radar image local point-surface contrast product ship detection method |
| CN113406625A (en) * | 2021-05-08 | 2021-09-17 | 杭州电子科技大学 | SAR image superpixel sliding window CFAR detection method |
| CN113781386A (en) * | 2021-05-06 | 2021-12-10 | 清华大学 | Ship detection method based on multi-source remote sensing image saliency fuzzy fusion |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016101279A1 (en) * | 2014-12-26 | 2016-06-30 | 中国海洋大学 | Quick detecting method for synthetic aperture radar image of ship target |
| US9389311B1 (en) * | 2015-02-19 | 2016-07-12 | Sandia Corporation | Superpixel edges for boundary detection |
| CN107067039A (en) * | 2017-04-25 | 2017-08-18 | 西安电子科技大学 | SAR image Ship Target quick determination method based on super-pixel |
-
2019
- 2019-02-26 CN CN201910142161.7A patent/CN109886218B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016101279A1 (en) * | 2014-12-26 | 2016-06-30 | 中国海洋大学 | Quick detecting method for synthetic aperture radar image of ship target |
| US9389311B1 (en) * | 2015-02-19 | 2016-07-12 | Sandia Corporation | Superpixel edges for boundary detection |
| CN107067039A (en) * | 2017-04-25 | 2017-08-18 | 西安电子科技大学 | SAR image Ship Target quick determination method based on super-pixel |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163450A (en) * | 2020-08-24 | 2021-01-01 | 中国海洋大学 | Based on S3High-frequency ground wave radar ship target detection method based on D learning algorithm |
| CN112183518A (en) * | 2020-09-25 | 2021-01-05 | 伏羲九针智能科技(北京)有限公司 | Vein target point automatic determination method, device and equipment |
| CN112164087B (en) * | 2020-10-13 | 2023-12-08 | 北京无线电测量研究所 | Super-pixel segmentation method and device based on edge constraint and segmentation boundary search |
| CN112164087A (en) * | 2020-10-13 | 2021-01-01 | 北京无线电测量研究所 | Super-pixel segmentation method and device based on edge constraint and segmentation boundary search |
| CN112528468A (en) * | 2020-11-20 | 2021-03-19 | 南京航空航天大学 | Label antenna reverse design method based on electromagnetic field SAR image processing |
| CN112766287B (en) * | 2021-02-05 | 2021-09-17 | 清华大学 | SAR image ship target detection acceleration method based on density examination |
| CN112766287A (en) * | 2021-02-05 | 2021-05-07 | 清华大学 | SAR image ship target detection acceleration method based on density examination |
| CN113781386A (en) * | 2021-05-06 | 2021-12-10 | 清华大学 | Ship detection method based on multi-source remote sensing image saliency fuzzy fusion |
| CN113781386B (en) * | 2021-05-06 | 2024-04-16 | 清华大学 | Ship detection method based on multi-source remote sensing image saliency fuzzy fusion |
| CN113406625A (en) * | 2021-05-08 | 2021-09-17 | 杭州电子科技大学 | SAR image superpixel sliding window CFAR detection method |
| CN113362293A (en) * | 2021-05-27 | 2021-09-07 | 西安理工大学 | SAR image ship target rapid detection method based on significance |
| CN113379694A (en) * | 2021-06-01 | 2021-09-10 | 大连海事大学 | Radar image local point-surface contrast product ship detection method |
| CN113379694B (en) * | 2021-06-01 | 2024-02-23 | 大连海事大学 | Radar image local point-to-face contrast product ship detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886218B (en) | 2022-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886218A (en) | SAR image Ship Target Detection method based on super-pixel statistics diversity | |
| CN110223302A (en) | A kind of naval vessel multi-target detection method extracted based on rotary area | |
| CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
| CN103871039B (en) | Generation method for difference chart in SAR (Synthetic Aperture Radar) image change detection | |
| CN104331886B (en) | Port region ship and warship detection method based on high resolution SAR image | |
| Gong et al. | Object detection based on improved YOLOv3-tiny | |
| CN103198475B (en) | Based on the total focus synthetic aperture perspective imaging method that multilevel iteration visualization is optimized | |
| CN106156758B (en) | A kind of tidal saltmarsh method in SAR seashore image | |
| Cai et al. | Guided attention network for object detection and counting on drones | |
| CN103425995B (en) | Hyperspectral image classification method based on region similarity low rank expression dimension reduction | |
| CN106600613B (en) | Improvement LBP infrared target detection method based on embedded gpu | |
| CN116563726A (en) | Remote sensing image ship target detection method based on convolutional neural network | |
| CN104268574A (en) | SAR image change detecting method based on genetic kernel fuzzy clustering | |
| Xue et al. | Rethinking automatic ship wake detection: state-of-the-art CNN-based wake detection via optical images | |
| CN105069459A (en) | Surface feature type extracting method for high-resolution SAR image | |
| CN110473255A (en) | A kind of ship bollard localization method divided based on multi grid | |
| CN112215832B (en) | SAR trail image quality evaluation and self-adaptive detection parameter adjustment method | |
| CN104077610B (en) | The method of the SAR image target recognition of two-dimension non linearity projection properties | |
| CN113935899A (en) | Ship plate image super-resolution method based on semantic information and gradient supervision | |
| CN116843906A (en) | Target multi-angle intrinsic feature mining method based on Laplace feature mapping | |
| CN116824485A (en) | Deep learning-based small target detection method for camouflage personnel in open scene | |
| Gao et al. | Scattering Characteristic-Aware Fully Polarized SAR Ship Detection Network Based on a Four-Component Decomposition Model | |
| CN114373135A (en) | Ship target detection method based on local significance characteristic measurement | |
| CN115496998A (en) | Remote sensing image wharf target detection method | |
| CN106204664B (en) | SAR Ship Target Detection method based on SAR-LARK feature |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |