CN102938150A - Synthetic aperture radar (SAR) image ship detection method based on self-adaptation sea clutter statistics - Google Patents

Abstract

The invention discloses a synthetic aperture radar (SAR) image ship separation detection method based on self-adaptation sea clutter statistics and mainly solves the problem that the effectiveness of sea surface ship detection in SAR image is low. The implementation process includes that images are subjected to threshold processing through an ostu threshold method to achieve sea surface separation; the images after sea surface separation are subjected to land rejection to obtain binary images and targets after the land rejection; the targets after the land rejection are subjected to boundary traverse to obtain boundary information; a self-adaptation background window is set by combining the original SAR images, the binary images after the land rejection and the target boundary information after the land rejection so as to complete background separation, and each target after the land rejection is subjected to target and background pixel statistics; and effective detection for ship targets is achieved through statistical information. Theoretical analysis and experimental result show that by means of the method, the accurate detection for SAR image sea surface ship targets can be achieved, and the method can be used for SAR image processing.

Description

SAR image Ship Detection based on self-adaptation sea clutter statistics

Technical field

The invention belongs to the Radar Technology field, the separation that relates to synthetic-aperture radar SAR image detects, and can realize the effective detection to sea Ship Target in the SAR image.

Background technology

Synthetic-aperture radar SAR is a kind of high-resolution microwave imaging radar, has the observation advantages such as remote, round-the-clock, round-the-clock.In recent years, utilize the SAR image that Ship Target is detected with the research that monitors and corresponding technological development and obtained great attention in the ocean remote sensing field, become one of very important application direction of SAR technology.

In SAR image Ship Detection, obtained studying widely and using based on two-parameter CFAR class with based on K-distributional class method.Two-parameter CFAR class methods background is assumed to be Gaussian distribution, by target setting window, protecting window, backdrop window and sliding step, to the detection of sliding of naval vessel class target in the image; Compare with two-parameter CFAR class methods, K-distributional class method adopts K-to distribute as extra large clutter distributed model, closer to the extra large clutter distributed model of reality, the method is by the calculation of parameter to image, match obtains the K-distributed model of background clutter, and the CFAR that is used for Ship Target detects, and overall K-location mode is estimated image parameter by view picture SAR image, once fitting background sea clutter distributes, and is applicable to middle low resolution SAR image; Local K-location mode is still by target setting window, protecting window, backdrop window and sliding step, finishes the detection of sliding of naval vessel class target in the image.

Above-mentioned two class methods have all adopted desirable distributed model, and simultaneously fixing window size and step-length setting has the following disadvantages it in reality detects:

Fixing distributed model can't accurate description SAR image in the Sea background clutter distribute;

2. complicated mathematical distribution model is unfavorable for Project Realization;

3. fixing window size can't be realized the accurate detection to different volumes size naval vessel;

4. when the unwanted target of appearance or background distributions were discontinuous in the backdrop window, method was unstable;

5. choosing of step-length has larger impact to counting yield and detection efficiency.

Summary of the invention

The object of the invention is to overcome the deficiency of above-mentioned prior art, a kind of SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics is proposed, to be implemented in the SAR image naval vessel, sea classification target is accurately detected, improve detection efficiency, make it more be conducive to Project Realization.

Realize that the object of the invention technical scheme is: by the pre-service to the SAR image, the adaptive background window is set, finishes the statistics to the different target background distributions, utilize this characteristic statistics result, non-naval vessel class target is rejected, realized naval vessel, sea classification target in the SAR image is effectively detected.Implementation step comprises as follows:

1) input sea SAR image obtains m * n dimension input picture matrix I;

2) utilize Otsu threshold method OSTU to find the solution the optimum thresholding G of image array I _Opg, with this optimum thresholding image array I is carried out threshold process, obtain the sea and reject rear binary map matrix I _e, finish the sea and separate;

3) reject on land;

3a) rear binary map matrix I is rejected on the sea _eIn each sea separate after target T _gCarry out connected volume traversal statistics, obtain each sea and separate rear target T _gBinary map matrix I after reject on the sea _eOn volume element information V _g, g=1,2 ..., N, wherein N is binary map matrix I after reject on the sea _eThe total number of middle target;

Be σ according to known radar resolution 3b) _a* σ _rWith the interval V～V' of actual naval vessel class objective body integration cloth,

Calculate the naval vessel at the distributed area V of SAR image upper volume unit _Min~ V _Max:

$V_{\min }=\frac{V}{{\mathrm{\σ}}_a\×{\mathrm{\σ}}_r},---\left(10\right)$

$V_{\max }=\frac{V^{\′}}{{\mathrm{\σ}}_a\×{\mathrm{\σ}}_r};---\left(11\right)$

3c) reject rear binary map matrix I in conjunction with the sea _eAnd rear target T is separated on each sea _gVolume information V _g, binary map matrix I after reject on the sea _eUpper to not meeting the interval V of volume distributed median _Min~ V _MaxTarget delete, obtain rejecting bianry image matrix I behind the land _t, reject bianry image matrix I behind the land _tIn each rejects target T behind land _k' expression, wherein, k=1,2 ..., N', N' reject bianry image matrix I behind the land _tTotal number of target behind the middle rejecting land;

4) successively to bianry image matrix I behind the rejecting land _tIn each rejects target T behind land _k' separate detection, obtain Ship Target:

4a) to bianry image matrix I behind the rejecting land _tTarget T behind the middle rejecting land _k' carry out Boundary Statistic, obtain its boundary information, according to boundary information, arrange and reject target T behind the land _k' adaptive background window matrix I _Bk

4b) utilize bianry image matrix I behind the rejecting land _t, adaptive background window matrix I _BkWith original SAR image array I, to target T behind the rejecting land _k' carry out background separation, and statistics is rejected target T behind the land respectively _k' object pixel and background pixel, obtain rejecting target T behind the land _k' the object pixel distribution p _Tk(x) and the background pixel distribution p _Bk(x);

4c) utilize the object pixel distribution p _Tk(x) calculate target T behind the rejecting land _k' pixel average μ _k, with the background pixel distribution p _Bk(x) distribution function is as a setting brought the CFAR detection formula into, tries to achieve and rejects target T behind the land _k' CFAR detection thresholding t _k, and judge: if μ _k〉=t _k, then judge and reject target T behind the land _k' be naval vessel class target, if μ _k＜t _k, then judge and reject target T behind the land _k' be non-naval vessel class target.

The present invention compared with prior art has the following advantages

The first, the present invention is based on the setting of adaptive background window, according to object boundary information, and Lookup protocol target background window size, thus avoided setting the impact that fixed background window, protecting window and target window bring;

Second, the present invention is based on the detection method of actual extra large clutter statistical model, can realize the actual count to extra large clutter distribution around the target in the image, utilize this statistical model to finish CFAR detection to target, thereby reduced because the detection error that fixing extra large clutter distributed model brings has improved detection validity;

The 3rd, the present invention passes through the target background isolation technics, realization detects the separation of target, when detecting low coverage or offshore target, avoid the impact that other targets are added up background distributions in the backdrop window, thereby overcome the classic method impact that other object pixels are estimated context parameter when context parameter is estimated;

The 4th, the present invention is to the statistics of target volume, can count accurately the pixel cell volume of distribution of same connection target, aspect the rejecting of land, utilize this statistical method complete rejecting can well be carried out in large volume land, thereby the impact of having avoided land that target detection is produced can realize the detection to the large volume Ship Target simultaneously;

Below in conjunction with accompanying drawing the present invention is described further.

Description of drawings

Fig. 1 is the process flow diagram of invention;

Fig. 2 is the original SAR image that emulation of the present invention is used;

Fig. 3 carries out result after separate on the sea with the present invention to Fig. 2;

Fig. 4 carries out result after reject on land with the present invention to Fig. 3;

Fig. 5 carries out result after the background separation with the present invention to naval vessel class target;

Fig. 6 carries out result after the background separation with the present invention to non-naval vessel class target;

Fig. 7 carries out the result that the naval vessel detects with the present invention to Fig. 2.

Embodiment

With reference to Fig. 1, implementation step of the present invention is as follows:

Step 1. input SAR image obtains m * n dimension image array I.

Step 2. adopts Otsu threshold method OSTU that the sea separation is separated on sea among the input picture matrix I.

OSTU processes and to be based on maximum between-cluster variance and to carry out threshold process, and it can be separated land, target and stronger extra large clutter etc. with the sea, thereby realizes that the sea separates, and Otsu threshold method OSTU concrete operations are divided into following step:

2a) establish that pixel grayscale is h among the image array I, h=0,1 ..., 255, the total number of pixels N among the computed image matrix I _u:

$N_u=\underset{h=0}{\overset{255}{\mathrm{\Σ}}}{n}_h,---\left(1\right)$

Wherein, n _hNumber of pixels for gray level h;

2b) calculate the probability P that gray level is the pixel appearance of h _h:

$P_h=\frac{n_h}{N_u},---\left(2\right)$

Wherein, P _h〉=0, and

2c) the total average gray μ of computed image matrix I:

$\mathrm{\μ}=\underset{h=0}{\overset{255}{\mathrm{\Σ}}}h{P}_h,---\left(3\right)$

2d) establish threshold value G gray level is divided into two groups of C ₀=0～G, C ₁=(G+1)～255, C ₀Represent background, C ₁Represent target, define respectively C ₀, C ₁The probability and the average that produce are as follows:

C ₀The probability ω that produces ₀:

${\mathrm{\ω}}_0=\underset{h=0}{\overset{G}{\mathrm{\Σ}}}{P}_h=\mathrm{\ω}\left(G\right),---\left(4\right)$

C ₁The probability ω that produces ₁:

${\mathrm{\ω}}_1=\underset{h=G+1}{\overset{255}{\mathrm{\Σ}}}{P}_h=1-\mathrm{\ω}\left(G\right),---\left(5\right)$

C ₀Average μ ₀:

${\mathrm{\μ}}_0=\frac{\underset{h=0}{\overset{G}{\mathrm{\Σ}}}h{P}_h}{{\mathrm{\ω}}_0}=\frac{\mathrm{\μ}\left(G\right)}{\mathrm{\ω}\left(G\right)},---\left(6\right)$

C ₁Average μ ₁:

${\mathrm{\μ}}_1=\frac{\underset{h=G+1}{\overset{255}{\mathrm{\Σ}}}h{P}_h}{{\mathrm{\ω}}_1}=\frac{\mathrm{\μ}-\mathrm{\μ}\left(G\right)}{1-\mathrm{\ω}\left(G\right)},---\left(7\right)$

Wherein, $\mathrm{\μ}\left(G\right)=\underset{h=0}{\overset{G}{\mathrm{\Σ}}}h{P}_h.$

Calculate through the inter-class variance after the threshold value G processing and be:

σ ²(G)＝ω ₀(μ ₀-μ) ²+ω ₁(μ ₁-μ) ²； （8）

2e) get respectively G=0～255, by poor σ between (8) formula compute classes side ²And note inter-class variance σ (G), ²Threshold value when (G) getting maximal value is optimal threshold G _Opg

2f) matrix I is set _e=I utilizes optimal threshold G _OpgImage is carried out threshold process:

$I_e(i,j)=\left\{\begin{array}{cc}255& I(i,j)\&GreaterEqual;G_{\mathrm{opg}}\\ 0& I(i,j)<G_{\mathrm{opg}}\end{array}\right.,---\left(9\right)$

Wherein, I _eBe binary map matrix after the rejecting of sea, i=1 ..., m, j=1 ..., n.

Through aforesaid operations, some weak pixels are removed in the image, thereby so that sea and non-sea-surface target separate.

Reject on step 3. land;

After the separating treatment of sea, binary map matrix I after reject on the sea _eIn only have I _eThe pixel of (i, j)=255 and I _eThe pixel of (i, j)=0, wherein I _eThe pixel of (i, j)=255 represents the targets such as land, extra large clutter and naval vessel.In realistic objective detected, the existence meeting on land caused very large impact to the detection of Ship Target, produces higher false-alarm probability.In order to realize the accurate detection to target, need to reject the land part, the concrete operations of rejecting are as follows:

3a) rear binary map matrix I is rejected on the sea _eIn each sea separate after target carry out connected volume traversal statistics, the concrete grammar step of connected volume traversal statistics is as follows:

Binary map matrix I after 3a1) reject on the input sea _e, sign matrix I is set _m, I _mFor m * n ties up full null matrix;

Binary map matrix I after 3a2) reject on the traversal sea _ePixel I _e(i, j), i=1 wherein ..., m, j=1 ..., if n is I _e(i, j)=255 and I _mThe rear target T of sea separation is then remembered in (i, j)=0 _g, current traversal coordinate (i, j) separates rear target T for the sea _gInitial coordinate (i _g, j _g), g=1 wherein, 2 ..., N, N are binary map matrix I after reject on the sea _eThe total number of target after separate on middle sea;

3a3) storehouse stack is set, and with coordinate (i _g, j _g) push on, make I _m(i _g, j _g)=1 arranges the sea and separates rear target T _gVolumetric parameter V _g, and initialization volumetric parameter V _g=0;

3a4) judge whether storehouse stack is empty, if storehouse stack is not empty, the stack top coordinate is popped, and is designated as (i', j '), makes V again _g=V _g+ 1, execution in step 3a5); If storehouse stack is empty, then records the sea and separate rear target T _gAnd volumetric parameter V _g, execution in step 3a6);

3a5) respectively to each pixel I in eight fields of stack top coordinate (i', j') _e(i ", j ") is by sequentially traveling through inspection clockwise or counterclockwise: if I _e(i ", j ")=255 and I _m(i ", j ")=0 is then with pixel I _e(i ", and the coordinate of j ") (i ", j ") puts into storehouse stack, and makes I _m(i ", j ")=1; After eight field pixel complete inspections of stack top coordinate (i', j') are complete, return step 3a4);

3a6) judge current traversal coordinate (i, j), if i=m and j=n, then connected volume traversal statistics finishes; Otherwise, return step 3a2).

By connected volume traversal statistical method, obtain each sea and separate rear target T _gBinary map matrix I after reject on the sea _eOn volume element information V _g, g=1,2 ..., N, wherein N is binary map matrix I after reject on the sea _eThe total number of middle target.

3b) the naval vessel classification is marked on the SAR image distribution unit interval and calculates, namely according to known radar resolution σ _a* σ _rWith the interval V～V' of actual naval vessel class objective body integration cloth, calculate the naval vessel at the distributed area V of SAR image upper volume unit _Min~ V _Max:

$V_{\min }=\frac{V}{{\mathrm{\&sigma;}}_a\&times;{\mathrm{\&sigma;}}_r},---\left(10\right)$

$V_{\max }=\frac{V^{\&prime;}}{{\mathrm{\&sigma;}}_a\&times;{\mathrm{\&sigma;}}_r};---\left(11\right)$

Wherein, σ _aBe SAR azimuth resolution, σ _rBe the SAR range resolution, V is actual naval vessel class target minimum volume, and V' is actual naval vessel class target maximum volume, V _MinBe naval vessel class target minimum volume unit number on the SAR image, V _MaxBe naval vessel class target maximum volume unit number on the SAR image;

3c) reject rear binary map matrix I in conjunction with the sea _eAnd rear target T is separated on each sea _gVolume information V _g, binary map matrix I after reject on the sea _eUpper to not meeting the interval V of volume distributed median _Min~ V _MaxTarget delete, concrete steps are as follows:

3c1) m * n dimension matrix I is set _t, and make I _t=I _e

3c2) to step 3a) in each sea separate after target T _gVolumetric parameter V _gJudge: if V _g＜V _MinOr V _gV _Max, then to I _tUpward all belong to target T _gPixel carry out zero setting; If V _Min≤ V _g≤ V _Max, then to I _tUpward all belong to target T _gPixel keep;

3c3) note I _tBe matrix I behind the rejecting land _t, I _tIn each target T _k' expression, the initial coordinate of each target is (i _k, j _k), wherein, k=1,2 ..., N', N' reject bianry image matrix I behind the land _tThe total number of middle target.

Step 4. target is separated detection;

After large volume land is rejected, obtain rejecting matrix I behind the land _t, I _tIn each rejects target T behind land _k' represented all kinds of targets that meet class objective body integration cloth interval, naval vessel, comprise non-naval vessel class target that strong sea clutter produces, non-naval vessel class target that small-sized island produce and naval vessel class target etc., in order to reject non-naval vessel class target, realization accurately detects naval vessel classification target, need to reject target target T behind the land to each _k' carry out the separation detection of background separation, concrete steps are as follows:

4a) according to matrix I behind the rejecting land _t, m * n dimension adaptive background window matrix I is set _Bk:

4a1) to matrix I behind the rejecting land _tIn each rejects target T behind land _k', with its initial coordinate (i _k, j _k) be the initial coordinate point, utilize and step 3a) identical method carries out the connected volume traversal, with target T behind the tense marker rejecting land _k' boundary coordinate:

$\left\{\begin{array}{c}{i}_{k\min =\underset{i\&Element;T_k^{\&prime;}}{\min }i}\\ i_{k\max }=\underset{i\&Element;T_k^{\&prime;}}{\max }i\\ j_{k\min =\underset{j\&Element;T_k^{\&prime;}}{\min }j}\\ j_{k\max =\underset{j\&Element;T_k^{\&prime;}}{\max }j}\end{array}\right.,---\left(12\right)$

Wherein, i is the row matrix coordinate, and j is the rectangular array coordinate, i _Kmin, i _KmaxRepresent each target T _k' bianry image matrix I after rejecting land _tThe minimum of middle row, maximum boundary row-coordinate, j _Kmin, j _KmaxRepresent respectively each and reject target T behind the land _k' bianry image matrix I after rejecting land _tThe minimum of middle row, maximum boundary row coordinate;

4a2) to target T behind each rejecting land _k' boundary coordinate do expansion processing:

$\left\{\begin{array}{c}{i}_{b\_k\min }=i_{k\min }-L_k\\ i_{b\_k\max }=i_{k\max }+L_k\\ j_{b\_k\min }=j_{k\min }-W_k\\ j_{b\_k\max }=j_{k\max }+W_k\end{array}\right.,---\left(13\right)$

Wherein, i _{B_kmin}, i _{B_kmax}Be respectively the border and enlarge rear minimum, maximum row coordinate, j _{B_kmin}, j _{B_kmax}Be respectively the border and enlarge rear minimum, maximum column coordinate, L _k=i _Kmax-i _Kmin, be target T behind the rejecting land _k' capable distribution length, W _k=j _Kmax-j _Kmin, be target T behind the rejecting land _k' column distribution length;

4a3) enlarge rear coordinate to adaptive background window matrix I according to the border _BkCarry out assignment, that is:

I _bk(i _{b_kmin}:i _{b_kmax},j _{b_kmin}:j _{b_kmax})＝1， （14）

Wherein, i _{B_kmin}: i _{B_kmax}Representative is from adaptive background window matrix I _BkI _{B_kmin}Row is to i _{B_kmax}OK, j _{B_kmin}: j _{B_kmax}Representative is from adaptive background window matrix I _BkJ _{B_kmin}Row are to j _{B_kmax}Row;

4b) in conjunction with bianry image matrix I behind SAR original image matrix I, the rejecting land _tWith adaptive background window matrix I _BkTarget is separated and statistics with background:

4b1) to matrix I behind the rejecting land _tCarry out normalized, obtain the image array I after the normalization _T1:

I _t1＝I _t./255， （15）

Wherein ./be matrix point except the operation;

4b2) according to the image array I after original SAR image I matrix and the normalization _T1, obtain target separation matrix I _T:

I _T＝I·I _t1， （16）

Wherein, the dot product for matrix operates;

4b3) according to original SAR image array I and adaptive background window matrix I _Bk, the adaptive background window is separated, obtain adaptive background window separation matrix I _BT:

I _BT＝I·I _bk， （17）

Wherein, the dot product for matrix operates;

4b4) according to adaptive background window separation matrix I _BTWith target separation matrix I _T, background is separated, obtain background separation matrix I _B:

I _B＝I _BT-I _T； （18）

4b5) respectively to target separation matrix I _TWith background separation matrix I _BMiddle pixel is added up, and obtains rejecting target T behind the land _k' the object pixel distribution p _Tk(x) and the background pixel distribution p _Bk(x);

4c) according to the pixel distribution p of target _Tk(x) and the background pixel distribution p _Bk(x), to target T behind the rejecting land _k' detect:

4c1) to target distribution p _Tk(x) carry out pixel average μ _kFind the solution:

μ _k＝mean(p _tk(x))， （19）

Wherein, mean () is the operation of averaging.

4c2) with the background distributions p of target _Bk(x) as the background distributions function of CFAR detection, according to the CFAR detection formula:

$1-P_{\mathrm{fa}}={\&Integral;}_0^{t_k}{p}_{\mathrm{bk}}\left(x\right)\mathrm{dx},---\left(20\right)$

Try to achieve CFAR thresholding t _k, P wherein _FaBe invariable false alerting;

4c3) with CFAR thresholding t _kTo target mean μ _kCarrying out thresholding judges: if μ _k〉=t _k, then judge and reject target T behind the land _k' be naval vessel class target; If μ _k＜t _k, then judge and reject target T behind the land _k' be non-naval vessel class target.

By above-mentioned steps, finished naval vessel classification target in the SAR image has accurately been detected.

Effect of the present invention can be illustrated by following emulation experiment:

1. simulated conditions

The present invention adopts SAR image as shown in Figure 3 to test, and wherein volume distributed median interval in naval vessel is made as 0 ~ 1000, and the naval vessel number is 11 in the image.

2. emulation content and result

Utilizing the inventive method that Fig. 2 is carried out naval vessel classification target detects:

2.1) with Fig. 2 input, obtain image array I, image array I is carried out Otsu threshold method OSTU process, obtain the sea and reject rear binary map matrix I _e, the sea separating resulting as shown in Figure 3.As can be seen from Figure 3, the targets such as land, naval vessel, extra large clutter are retained, and the sea is separated, thereby has reached good sea separating effect.

2.2) rear binary map matrix I is rejected on the sea _eCarry out volume statistics, and large area target is wherein rejected, obtain the target binary map matrix I after reject on land _t, the result is rejected as shown in Figure 4 in land.As can be seen from Figure 4, large-area land is disallowable, and what retain is to meet all kinds of targets that the naval vessel distributed area requires.

2.3) reject rear image array I according to original image matrix I, land _tWith adaptive background window matrix I _BkBackground is separated, and adds up its distribution, the result as shown in Figure 5 and Figure 6, wherein, Fig. 5 is naval vessel classification target separating resulting, Fig. 6 is non-naval vessel class target separating resulting.Can find out from Fig. 5, Fig. 6, target and background is well separated, naval vessel class target pixel distribution concentrates on high pixel range simultaneously, and low pixel range utilized this character to finish non-naval vessel classification target is rejected during non-naval vessel class target pixel distribution concentrated on.

2.4) calculate target mean μ according to the distribution of target and background _kWith background CFAR thresholding t _k, each target after separating in the SAR image being carried out non-naval vessel class target reject, the result is as shown in Figure 7.As can be seen from Figure 7, by the inventive method, realized naval vessel classification target in the image is accurately detected.

Performance index such as the table 1 of testing result among Fig. 7:

The performance index contrast that table 1 detects

The naval vessel actual number	Undetected number	The correct number that detects	The false-alarm number
				11	0	11	0

As can be seen from Table 1, detection method of the present invention can realize naval vessel classification target in the SAR image is accurately detected.

Claims (7)

1. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics comprises the steps:

1) input sea SAR image obtains m * n dimension image array I;

2) utilize Otsu threshold method OSTU to find the solution the optimum thresholding G of image array I _Opg, with this optimum thresholding image array I is carried out threshold process, obtain the sea and reject rear binary map matrix I _e, finish the sea and separate;

3) reject on land;

3a) rear binary map matrix I is rejected on the sea _eIn each sea separate after target T _gCarry out connected volume traversal statistics, obtain each sea and separate rear target T _gBinary map matrix I after reject on the sea _eOn volume element information V _g, g=1,2 ..., N, wherein N is binary map matrix I after reject on the sea _eThe total number of middle target;

Be σ according to known radar resolution 3b) _a* σ _rWith the interval V～V' of actual naval vessel class objective body integration cloth, calculate the naval vessel at the distributed area V of SAR image upper volume unit _Min~ V _Max:

$V_{\min }=\frac{V}{{\mathrm{\&sigma;}}_a\&times;{\mathrm{\&sigma;}}_r},---\left(10\right)$

$V_{\max }=\frac{V^{\&prime;}}{{\mathrm{\&sigma;}}_a\&times;{\mathrm{\&sigma;}}_r};---\left(11\right)$

3c) reject rear binary map matrix I in conjunction with the sea _eAnd rear target T is separated on each sea _gVolume information V _g, binary map matrix I after reject on the sea _eUpper to not meeting the interval V of volume distributed median _Min~ V _MaxTarget delete, obtain rejecting bianry image matrix I behind the land _t, reject bianry image matrix I behind the land _tIn each rejects target T behind land _k' expression, wherein, k=1,2 ..., N', N' reject bianry image matrix I behind the land _tTotal number of target behind the middle rejecting land;

4) successively to bianry image matrix I behind the rejecting land _tIn each rejects target T behind land _k' separate detection, obtain Ship Target:

4a) to bianry image matrix I behind the rejecting land _tTarget T behind the middle rejecting land _k' carry out Boundary Statistic, obtain its boundary information, according to boundary information, arrange and reject target T behind the land _k' adaptive background window matrix I _Bk

4b) utilize bianry image matrix I behind the rejecting land _t, adaptive background window matrix I _BkWith original SAR image array I, to target T behind the rejecting land _k' carry out background separation, and statistics is rejected target T behind the land respectively _k' object pixel and background pixel, obtain rejecting target T behind the land _k' the object pixel distribution p _Tk(x) and the background pixel distribution p _Bk(x);

4c) utilize the object pixel distribution p _Tk(x) calculate target T behind the rejecting land _k' pixel average μ _k, with the background pixel distribution p _Bk(x) distribution function is as a setting brought the CFAR detection formula into, tries to achieve and rejects target T behind the land _k' CFAR detection thresholding t _k, and judge: if μ _k〉=t _k, then judge and reject target T behind the land _k' be naval vessel class target, if μ _k＜t _k, then judge and reject target T behind the land _k' be non-naval vessel class target.

2. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 3a) described the sea is rejected after binary map matrix I _eCarry out connected volume traversal statistics, carry out as follows:

Binary map matrix I after 3a1) reject on the input sea _e, sign matrix I is set _m, I _mFor m * n ties up full null matrix;

Binary map matrix I after 3a2) reject on the traversal sea _ePixel I _e(i, j), i=1 wherein ..., m, j=1 ..., if n is I _e(i, j)=255 and I _mThe rear target T of sea separation is then remembered in (i, j)=0 _g, current traversal coordinate (i, j) separates rear target T for the sea _gInitial coordinate (i _g, j _g), g=1 wherein, 2 ..., N, N are binary map matrix I after reject on the sea _eThe total number of target after separate on middle sea;

3a3) storehouse stack is set, and with coordinate (i _g, j _g) push on, make I _m(i _g, j _g)=1 arranges the sea and separates rear target T _gVolumetric parameter V _g, and initialization volumetric parameter V _g=0;

3a4) judge whether storehouse stack is empty, if storehouse stack is not empty, the stack top coordinate is popped, and is designated as (i', j '), makes V again _g=V _g+ 1, execution in step 3a5); If storehouse stack is empty, then records the sea and separate rear target T _gAnd volumetric parameter V _g, execution in step 3a6);

3a5) respectively to each pixel I in eight fields of stack top coordinate (i', j') _e(i ", j ") is by sequentially traveling through inspection clockwise or counterclockwise: if I _e(i ", j ")=255 and I _m(i ", j ")=0 is then with pixel I _e(i ", and the coordinate of j ") (i ", j ") puts into storehouse stack, and makes I _m(i ", j ")=1; After eight field pixel complete inspections of stack top coordinate (i', j') are complete, return step 3a4);

3a6) judge current traversal coordinate (i, j), if i=m and j=n, then connected volume traversal statistics finishes; Otherwise, return step 3a2).

3. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 3c) described after reject on the sea binary map matrix I _eUpper to not meeting the interval V of volume distributed median _Min~ V _MaxTarget delete, carry out as follows:

3c1) m * n dimension matrix I is set _t, and make I _t=I _e

3c2) to step 3a) in each sea separate after target T _gVolumetric parameter V _gJudge: if V _g＜V _MinOr V _gV _Max, then to I _tUpward all belong to target T _gPixel carry out zero setting; If V _Min≤ V _g≤ V _Max, then to I _tUpward all belong to target T _gPixel keep;

3c3) note I _tBe matrix I behind the rejecting land _t, I _tIn each target T _k' expression, the initial coordinate of each target is (i _k, j _k), wherein, k=1,2 ..., N', N' reject bianry image matrix I behind the land _tThe total number of middle target.

4. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 4a) the adaptive background window matrix I of described Offered target _Bk, carry out as follows:

4a1) the complete zero adaptive background window matrix I of m * n dimension is set _Bk

4a2) to matrix I behind the rejecting land _tIn each rejects target T behind land _k', with its initial coordinate (i _k, j _k) be the initial coordinate point, utilize and step 3a) identical method carries out the connected volume traversal, with target T behind the tense marker rejecting land _k' the border:

$\left\{\begin{array}{c}{i}_{k\min }=\underset{i\&Element;{T_k}^{\&prime;}}{\min i}\\ i_{k\max }=\underset{i\&Element;{T_k}^{\&prime;}}{\max i}\\ j_{k\min }=\underset{j\&Element;{T_k}^{\&prime;}}{\min j}\\ j_{k\max }=\underset{j\&Element;{T_k}^{\&prime;}}{\max j}\end{array}\right.,---\left(12\right)$

Wherein, i is the row matrix coordinate, and j is the rectangular array coordinate, i _Kmin, i _KmaxRepresent respectively each target T _k' bianry image matrix I after rejecting land _tThe minimum of middle row, maximum boundary row-coordinate, j _Kmin, j _KmaxRepresent respectively each target T _k' bianry image matrix I after rejecting land _tThe minimum of middle row, maximum boundary row coordinate;

4a3) to each mark of order T _k' boundary coordinate do expansion processing, obtain target T _k' the rear coordinate of border expansion:

$\left\{\begin{array}{c}{i}_{b\_k\min }=i_{k\min }-L_k\\ i_{b\_k\max }=i_{k\max }+L_k\\ j_{b\_k\min }=j_{k\min }-W_k\\ j_{b\_k\max }=j_{k\max }+W_k\end{array}\right.,---\left(13\right)$

Wherein, i _{B_kmin}, i _{B_kmax}Be respectively the border and enlarge rear minimum, maximum row coordinate, j _{B_kmin}, j _{B_kmax}Be respectively the border and enlarge rear minimum, maximum column coordinate, L _k=i _Kmax-i _Kmin, be target T behind the rejecting land _k' capable distribution length, W _k=j _Kmax-j _Kmin, be target T behind the rejecting land _k' column distribution length;

4a4) m * n is set and ties up full null matrix I _Bk, and after enlarging in conjunction with the border coordinate to I _BkCarry out assignment, obtain adaptive background window matrix I _Bk:

I _bk(i _{b_kmin}:i _{b_kmax},j _{b_kmin}:k _{b_kmax})＝1 （14）

Wherein, i _{B_kmin}: i _{B_kmax}Representative is from matrix I _BkI _{B_kmin}Row is to i _{B_kmax}OK, j _{B_kmin}: j _{B_kmax}Representative is from matrix I _BkJ _{B_kmin}Row are to j _{B_kmax}Row.

5. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 4b) describedly obtain rejecting target T behind the land _k' the object pixel distribution p _Tk(x) and the background pixel distribution p _Bk(x), carry out as follows:

4b1) to matrix I behind the rejecting land _tCarry out normalized, obtain the image array I after the normalization _T1:

I _t1＝I _t./255， （15）

Wherein ./be matrix point except the operation;

4b2) according to the image array I after original SAR image I matrix and the normalization _T1, target is separated, obtain target separation matrix I _T:

I _T＝I·I _t1， （16）

Wherein, the dot product for matrix operates;

4b3) according to original SAR image array I and adaptive background window matrix I _Bk, the adaptive background window is separated, obtain adaptive background window separation matrix I _BT:

I _BT＝I·I _bk， （17）

Wherein, the dot product for matrix operates;

4b4) according to adaptive background window separation matrix I _BTWith target separation matrix I _T, background is separated, obtain background separation matrix I _B:

I _B＝I _BT-I _T； （18）

4b5) respectively to target separation matrix I _TWith background separation matrix I _BMiddle pixel is added up, and obtains rejecting target T behind the land _K' the object pixel distribution p _Tk(x) and the background pixel distribution p _Bk(x).

6. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 4c) the described object pixel distribution p of utilizing _Tk(x) calculate target T behind the rejecting land _k' pixel average μ _k, undertaken by following formula:

μ _k＝mean(p _tk(x))， （19）

Wherein, mean () is the operation of averaging.

7. the SAR image naval vessel method for separating and detecting based on self-adaptation sea clutter statistics according to claim 1, wherein step 4c) described calculating rejects target T behind the land _k' CFAR detection thresholding t _k, be the background distributions p with target _Bk(x) as the background distributions function of CFAR detection formula, according to following CFAR detection formula, computing constant false-alarm thresholding t _k:

$1-P_{\mathrm{fa}}={\&Integral;}_0^{t_k}{p}_{\mathrm{bk}}\left(x\right)\mathrm{dx},---\left(20\right)$

P wherein _FaBe invariable false alerting.

Images