CN103971127A - Forward-looking radar imaging sea-surface target key point detection and recognition method - Google Patents

Abstract

The invention discloses a forward-looking radar imaging sea-surface target key point detection and recognition method. The method includes the steps that radar echo data are quantized into a gray scale image; a region of interest is extracted from the gray scale image, partition is performed, and a target region partition image is obtained; the radar echo data and the region of interest of the image are used, and a target region peak point information matrix is obtained; information merging is conducted on the target region partition image and the target region peak point information matrix, the number K of peak points in a target region in a merging result is obtained through counting, the previous K peak points are selected and serve as target effective peak points and form a target effective peak point image in a binaryzation mode, target axial features are extracted, and a target position is determined; the center of gravity of target energy is calculated and serves as the target key point. According to forward-looking radar target features, multiple mode identification methods are used comprehensively, the inherent features of a target can be reserved, meanwhile interference factors such as artifacts and side lobes can be restrained, and the recognition accuracy and the positioning precision of radar imaging sea surface target key points are improved.

Description

The recognition methods of a kind of forward-looking radar imaging sea-surface target critical point detection

Technical field

The invention belongs to target detection, mode identification technology, be specifically related to the recognition methods of a kind of forward-looking radar imaging sea-surface target critical point detection, the method can effectively suppress the disturbing factor such as artifact, secondary lobe in retaining target inherent characteristic, improves recognition correct rate and positioning precision to radar imagery sea-surface target key point.

Background technology

The Ship Target Detection identification that utilizes radar imagery technology to realize is the gordian technique of many civilian and military fields, is widely used in monitoring, the field such as military.In forward-looking radar echo, strong echoed signal means that detector searches strong scattering point at this place.Strong scattering point is normally caused by two reflecting bodys in Ship Target and corner reflector.These reflection parts are distributed on whole Ship Target, present different intensity sizes with the difference of azimuth of target.But, normal radar transmitting linear FM signal, because the two-dimensional frequency supporting domain of imaging system is limited, makes synthetic-aperture radar (Synthetic Aperture Radar, SAR) impulse response function distance to orientation to being sinc function, cause sidelobe level very high.Due to the processing window limited length of radar data, and there is phase error in radar return data, form secondary lobe, cause the noise of multiplication, and produce and interfere with near scatterer, very large on picture quality impact.The existence of secondary lobe causes the target image obtaining in forward sight imaging to have artifact, in intensity, will be weaker than real target point, and this is also the basis of the identification of target critical point and hi-Fix.False target and real goal can be distinguished, also need to consider the forward sight imaging under different points of view to target, whether the spacing of false target and real goal in image, there is the situation of overlapping interference.Therefore, be necessary before forward-looking radar target's feature-extraction, detection and Identification, original radar echo signal data are carried out to pre-service, to reduce noise effect, improve the signal to noise ratio (S/N ratio) of image, outstanding target signature information, utilizes target signature identification target, thereby improve the probability of target critical point identification, and utilize the energy barycenter of target to determine the key point of target.

In existing disclosed document, it is to realize by the processing to original radar echo signal mostly that the forward-looking radar imaging sea-surface target of mentioning detects recognition methods.But owing to restricted by SAR image-forming mechanism, target slice image is subject to the impact of the factors such as targeted attitude, background characteristics value and sensor imaging attitude, show higher changeableness, thereby easily cause the interference to recognition result, there is erroneous judgement and misjudgement.The people such as Zhang Hong have proposed the recognition methods of based target peak value feature in " identification of High Resolution SAR Images target ", are 20～40 the span of taking experimental data to add up to provide for choosing of peak point number.It is determined for peak point number, does not provide an effective criterion and method, can only be according to experience in engineering application.In existing open source literature, maximum entropy dividing method has advantages of good stability, but is easily subject to background interference, and segmentation result exists false target, gained target information inaccurate.

Summary of the invention

The target the present invention is directed under complex environment is extracted problem, proposes the recognition methods of a kind of forward-looking radar imaging sea-surface target critical point detection, specifically comprises:

(1) original radar two dimension echo data is quantified as to 2-D gray image data;

(2) 2-D gray image step (1) being obtained utilizes the method for based target physical dimension and degree of confidence to carry out region of interest extraction, obtains target area gray level image;

(3) use maximum entropy to cut apart to target area gray level image, must arrive target area and cut apart image;

(4) utilize radar two dimension echo data and target area gray level image, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix;

(5) image is cut apart in target area and target area peak point information matrix carries out information fusion, in statistics fusion results, peak point number K in target area, counts out as effective peak;

(6) peak point in the peak point information matrix of target area is arranged by size, chosen a front K peak point as target effective peak point, target area peak point information matrix two-value is turned to target effective peak point image;

(7) extract the axial feature of target in target effective peak point image, get rid of false-alarm point and disturb, determine target location;

(8) utilize target location and target energy center of gravity, determine target critical point.

Further, described step (1) specifically comprises:

(1.1) the floating point values threshold value that selection quantizes, establishing floating point values upper threshold is L _max, under threshold value, be limited to L _min, wherein:

$L_{\min }=\mathrm{Totalpix}*\frac{\min T}{\mathrm{Totalpix}}$

L _max＝N*(TLength*Margin) ²+L _min，if L _max＜Totalpix

L _max＝Totalpix,if L _max＞Totalpix

N is the target maximum number that may contain in background, Totalpix is the number of pixels of original image, TLength is the pixel count of target length in image, and Margin is surplus to ensure that target can complete demonstration, and minT is that the minimum of target two dimension echo data may floating point values;

(1.2) selective value is L respectively _maxfloating point values and value for L _minfloating point values as threshold value Level255 and Level0; To each data point in original radar two dimension echo data, if being greater than Level255, floating point values gives gray-scale value 255, be less than Level0 and give gray-scale value 0, to being worth at L _maxwith L _minbetween floating point values carry out linear interpolation, determine its gray-scale value, the formula of linear interpolation is as follows:

$g(x,y)=\left\{\begin{array}{cc}0,& \mathrm{if}(f(x,y)<\mathrm{Level}0)\\ \frac{f(x,y)-\mathrm{Level}0}{\mathrm{Level}255-\mathrm{Level}0}*255,& \mathrm{if}(\mathrm{Level}0\&le;f(x,y)\&le;\mathrm{Level}255)\\ 255,& \mathrm{if}(f(x,y)>\mathrm{Level}255)\end{array}\right.$

Wherein f (x, y) is the floating-point numerical value that data point (x, y) is located radar two dimension echo data, and g (x, y) is gray-scale value corresponding to this point (x, y) after linear interpolation.

Further, described step (2) specifically comprises:

(2.1), first according to target sizes and imaging resolution, using the interval under the state of the normal course of target as constraint, determine region of interest window length and width; Investigate the region of interest window centered by s point, in window, the statistic of pixel value has average μ _swith center of gravity G _s:

${\mathrm{\&mu;}}_s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}g(x_i,y_i)$

$\stackrel{\&OverBar;}{x}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}x\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$\stackrel{\&OverBar;}{y}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}y\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$G_s=(\stackrel{\&OverBar;}{x},\stackrel{\&OverBar;}{y})$

Wherein, n is the number of pixel in the window of region of interest, and g (x, y) is the gray-scale value that in the window of region of interest, (x, y) locates pixel, and Ω is the contained region of region of interest window;

(2.2) ask for the degree of confidence μ of the region of interest window centered by s point _s, and center of gravity G in window _swith window center coordinate O _sdistance d (G _s, O _s), wherein:

ρ _s＝μ _s/[d(G _s,O _s)+1]

d (G _s, O _s) be center of gravity G in window _swith window center coordinate O _sbetween Euclidean distance, wherein x, y is the row and column of presentation video respectively;

(2.3) in 2-D gray image, determine region of interest window according to following principle, as target area gray level image:

If be background in the region of interest window centered by s point, due to background pixel value difference not quite and all lower, now apart from d (G _s, O _s) ≈ 0, degree of confidence ρ _s≈ μ _ssuitable with background pixel value average, now Bu Shi region of interest, region of interest window region;

If a part is that a target part is background, now μ in window in the region of interest window centered by s point _sincrease, but center of gravity G in while window _sbe partial to the region of high pixel value, d (G _s, O _s) also increase, until region of interest window comprises whole target, be defined as region of interest;

If comprised whole target in the region of interest window centered by s point, now μ _sreach maximum, maximal value is relevant to the pixel value of target, simultaneously center of gravity G in window _swith window center O _sdistance reduces; When apart from d (G _s, O _s) degree of confidence ρ when ≈ 0 _sreach maximum, simultaneously ρ _sfor local maximum, now window region in region of interest is region of interest.

Further, described step (3) specifically comprises:

(3.1) be identified for segmentation threshold Th that target area gray level image is cut apart, segmentation threshold Th is target entropy Entropy _owith background entropy Entropy _bthe maximal value of sum, wherein:

Background entropy ${\mathrm{Entropy}}_B=-\underset{i=0}{\overset{\mathrm{Th}}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{P_t}\ln \frac{p\left(i\right)}{P_t}$

Target entropy ${\mathrm{Entropy}}_O=-\underset{i=\mathrm{Th}+1}{\overset{m}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{H_t}\ln \frac{p\left(i\right)}{H_t}$

P (i) represents that corresponding grey scale level is the probability size of i, the maximum gray scale of m presentation video, P _tand H _trepresent respectively background and target in image pixel grey scale distribute probability and;

(3.2) according to segmentation threshold Th, target area gray level image is cut apart, must be arrived target area and cut apart image O (x, y), be specially:

If the entropy of certain gray level is greater than Th, think that this gray level is target; If the entropy of certain gray level is less than Th, think that this gray level is background, cuts apart image O (x, y) thereby must arrive target area.

Further, described step (4) is specially:

According to radar two dimension echo data and target area gray level image, adopt first order difference method, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix G (x, y), wherein:

$G(x,y)=\left\{\begin{array}{cc}255,& \mathrm{if}(f(x,y)-f(x,y+1)>0,f(x,y)-f(x,y-1)>0)\\ 0,& \mathrm{else}\end{array}\right.$

(x, y) represents the point in two-dimentional echoed signal, x represent distance to, y represent orientation to, note radar two dimension echo floating point values is f (x, y).

Further, described step (5) is specially:

Image O (x, y) is cut apart to and target area peak point information matrix G (x, y) carries out information fusion in target area, obtain the image R (x, y) after merging;

R(x,y)＝G(x,y)·O(x,y)/255

Image R (x, y) after fusion is 0/255 bianry image, and the pixel that in image, gray level is 255 is target area and cuts apart peak point in objective area in image, adds up these peak point numbers K

$K=C_{\mathrm{ount}}\left\{(x_i,y_i)\right|\left.\begin{array}{c}(x_i,y_i)\&Element;R(x,y)\\ p(x_i,y_i)=255\end{array}\right\}$

Wherein, p (x _i, y _i) be (x in the image R (x, y) after merging _i, y _i) the gray-scale value size located.

Further, described step (6) specifically comprises:

(6.1) peak point in the peak point information matrix of target area is sorted by size, choose the effective peak point characteristic information of front K maximum peak point as target;

(6.2) the effective peak point characteristic information of target is carried out to binary conversion treatment, by this K extreme point place pixel assignment gray level 255, rest of pixels point assignment is 0, obtains target effective peak point image.

Further, described step (7) specifically comprises:

(7.1) in target effective peak point image to representing that k candidate target point utilizes least square fitting, search out a straight line that makes residual sum of squares (RSS) minimum, i.e. the axial straight line of target, this straight line is identical with the diagonal of target;

(7.2) using the length of target, width, Diagonal Dimension as constraint, taking the target rectangle of target length, wide constraint as window, allow the diagonal line of window along the axial rectilinear movement of target, find on the position on axial straight line and make maximum the window's position of contained peak point number in window, the corresponding position of window is target location;

(7.3) the peak point information flag outside window is false-alarm point, and in window, contained peak point information has been determined the position of target.

Further, in described step (7.1), the axial straight line of calculating target is specially:

To target pixel points set V, estimate the parameter k of straight line y=kx+b, b, makes residual sum of squares (RSS) minimum, wherein (x _i, y _i) ∈ V.

Further, described step (8) is specially:

Ask for the grey scale centre of gravity of target area set it as the energy barycenter of target area, be the key point of target, wherein to ask for formula as follows for the grey scale centre of gravity of target area:

$\stackrel{\&OverBar;}{X}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}x\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

$\stackrel{\&OverBar;}{Y}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}y\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

F (x, y) is the floating point values that in radar two dimension echo data, (x, y) locates, and Τ is target area.

In general, compared with prior art, the present invention has following beneficial effect to the above technical scheme of conceiving by the present invention:

(1) when utilizing radar return data peaks dot information to reject false target information, utilizing OTSU image to cut apart constraint effective peak counts out, choose in the self-adaptation of engineering applications peak point number K thereby realize, make the more standardization of result obtaining, can not change because of operator's difference;

(2) in the image quantization stage, utilize the relevant informations such as target and background size, set up the calculation criterion to quantization threshold, thereby made radar two dimension echo data show and to process with the form of image, for the image conversion processing of radar two dimension echo data provides Data support;

(3) according to radar image distance to orientation to imaging characteristics, utilize radar bearing to maximum point information determine target area, thereby on the basis that does not affect performance, realized the information extraction of target maximum value, local maximum extracting method has reduced calculated amount relatively;

(4) utilize the axial feature of target, the location of realize target and false-alarm point are got rid of, and make target location more accurate;

(5) proposed a kind of target region of interest extracting method that utilizes target physical dimension and degree of confidence, the region of interest of extraction is more accurate, has reduced the operand of subsequent treatment;

(6) treatment scheme of forward-looking radar imaging sponge target critical spot check detection identifying method to forward-looking radar echoed signal that the present invention proposes, the results show the validity of this flow process.

In sum, the present invention is according to forward-looking radar target property, Integrated using various modes recognition methods, can in retaining target inherent characteristic, suppress the disturbing factor such as artifact, secondary lobe, improve recognition correct rate and the positioning precision of radar imagery sea-surface target key point.Carry out target detection identification by this method, target is can discrimination high.

Brief description of the drawings

Fig. 1 is the overview flow chart of forward-looking radar imaging sea-surface target critical point detection of the present invention recognition methods;

Fig. 2 is the image after the quantification of original radar two dimension echo data in one embodiment of the invention;

Fig. 3 is the peak value characteristic pattern of original radar two dimension echo data;

Fig. 4 is that after original radar two dimension echo data in Fig. 2 is quantized, image segmentation result shows;

Fig. 5 is image object effective peak hum pattern after quantizing in original radar two dimension echo data in Fig. 2;

Fig. 6 is that in Fig. 2, original radar two dimension echo data quantizes the axial feature constraint result of target figure in rear image;

Fig. 7 is that in Fig. 2, original radar two dimension echo data quantizes target critical point location result in rear image.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.In addition,, in each embodiment of described the present invention, involved technical characterictic just can combine mutually as long as do not form each other conflict.

When the object of the invention is to retain target inherent characteristic, suppress the disturbing factor such as artifact, secondary lobe, improve recognition correct rate and the positioning precision of radar imagery sea-surface target key point.But in forward-looking radar echo, strong echoed signal means that detector searches strong scattering point at this place.Strong scattering point is normally caused by two reflecting bodys in Ship Target and corner reflector.These reflection parts are distributed on whole Ship Target, present different intensity sizes with the difference of azimuth of target.But, normal radar transmitting linear frequency modulation (Linear Frequency Modulation, LFM) signal, because the two-dimensional frequency supporting domain of imaging system is limited, the impulse response function that makes SAR distance to orientation to being sinc function, sidelobe level is very high.Because secondary lobe may form the noise of multiplication, and produce and interfere with near scatterer, very large on picture quality impact.The existence of secondary lobe causes the target image obtaining in forward sight imaging to have artifact, in intensity, will be weaker than real target point, and this is also the basis of target identification and hi-Fix.False target and real goal can be distinguished, also need to consider the forward sight imaging under different points of view to target, whether the spacing of false target and real goal in image, there is the situation of overlapping interference.The target property and the Sea background characteristic that the present invention is based in radar image are made hypothesis estimation.First, after utilizing radar original echo data and processing, carry out information fusion through the image of over-segmentation, obtain the peak value characteristic information of target area; Then, utilize the axial feature knowledge of target to get rid of interference, identification target.Finally, according to the energy barycenter localizing objects key point of target.

The present invention, first by the feature extraction of based target peak value, obtains the peak value characteristic pattern of target; Through quantizing to become image, after waiting resolution adjustment, use maximum entropy to cut apart original echo data, obtain cutting apart image; By the peak value characteristic pattern of gained with cut apart image and merge and obtain peak point number threshold value, can obtain effective target peak information, utilize axial feature and the energy barycenter of target, identify target, locator key point.Experimental result shows, when algorithm can suppress to retain target inherent characteristic in literary composition, reduces various disturbing factors, improves recognition correct rate and the positioning precision of target critical point.

The invention provides the recognition methods of a kind of forward-looking radar imaging sea-surface target critical point detection, as shown in Figure 1, the method detailed process is its overall procedure:

(1) original radar two dimension echo data is quantified as to 2-D gray image data

First original radar two dimension echo data is quantized, be transformed into the 2-D gray image data that the gray-scale value in 0～255 scope can carry out Digital Image Processing.The method that original radar two dimension echo data is mapped as to 256 grades of gray level images is: the floating point values of original radar two dimension echo data is sorted from small to large; Select the floating point values threshold value quantizing according to following principle, suppose that floating point values upper threshold is L _max, under threshold value, be limited to L _min, have

$L_{\min }=\mathrm{Totalpix}*\frac{\min T}{\mathrm{Totalpix}}$

L _max＝N*(TLength*Margin) ²+L _min，if L _max＜Totalpix

L _max＝Totalpix,if L _max＞Totalpix

Wherein, N is the target maximum number that may contain in background, the number of pixels that Totalpix is original image, TLength is the pixel count of target length in image, Margin is surplus, ensures the complete demonstration of target energy, and minT is that the minimum of target two dimension echo data may floating point values.

Selective value is L respectively _maxfloating point values and value for L _minfloating point values as threshold value Level255 and Level0; To each pixel, if being greater than Level255, floating point values gives gray-scale value 255, be less than Level0 and give gray-scale value 0, to being worth at L _maxwith L _minbetween floating point values carry out linear interpolation, determine its gray-scale value.The formula of linear interpolation is as follows

$g(x,y)=\left\{\begin{array}{cc}0,& \mathrm{if}(f(x,y)<\mathrm{Level}0)\\ \frac{f(x,y)-\mathrm{Level}0}{\mathrm{Level}255-\mathrm{Level}0}*255,& \mathrm{if}(\mathrm{Level}0\&le;f(x,y)\&le;\mathrm{Level}255)\\ 255,& \mathrm{if}(f(x,y)>\mathrm{Level}255)\end{array}\right.$

Wherein f (x, y) is the floating-point numerical value that pixel (x, y) is located radar two dimension echo data, and g (x, y) is gray-scale value corresponding to this point after linear interpolation, and the image after quantification as shown in Figure 2.

(2) 2-D gray image step (1) being obtained utilizes the method for based target physical dimension and degree of confidence to carry out region of interest extraction, obtains target area gray level image

First according to target sizes and imaging resolution, using the interval under the state of the normal course of target as constraint, determine suitable region of interest window length and width.Investigate the region of interest window centered by s point, in window, the statistic of pixel value has average μ _swith center of gravity G _s.

${\mathrm{\&mu;}}_s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}g(x_i,y_i)$

$\stackrel{\&OverBar;}{x}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}x\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$\stackrel{\&OverBar;}{y}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}y\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$G_s=(\stackrel{\&OverBar;}{x},\stackrel{\&OverBar;}{y})$

Wherein, n is the number of pixel in the window of region of interest, and g (x, y) is the gray-scale value that (x, y) locates pixel, and Ω is the contained region of region of interest window.

By μ _swhether be an evaluation criterion of region of interest as decision window, μ _sin the local window of higher explanation, pixel value is larger, is more likely target; Meanwhile, due to hope target lock-on the center in region of interest, center of gravity G in window _swith window center coordinate O _sdistance d (G _s, O _s) as another evaluation criterion.The region of interest degree of confidence of definition centered by s point as:

ρ _s＝μ _s/[d(G _s,O _s)+1]

Apart from d (G _s, O _s) be chosen for center of gravity G in window _swith window center coordinate O _sbetween Euclidean distance, wherein x, y is the row and column of presentation video respectively:

$d(G_s,O_s)=\sqrt{{(x_G-x_O)}^2+{(y_G-y_O)}^2}$

To three kinds of regions in 2-D gray image, process respectively:

If be background in the region of interest window centered by s point, due to background pixel value difference not quite and all lower, now apart from d (G _s, O _s) ≈ 0, degree of confidence ρ _s≈ μ _ssuitable with background pixel value average, now Bu Shi region of interest, region of interest window region;

If a part is that a target part is background, now μ in window in the region of interest window centered by s point _sincrease, but center of gravity G in while window _sbe partial to the region of high pixel value, d (G _s, O _s) also increase, until region of interest window comprises whole target, be defined as region of interest;

If comprised whole target in the region of interest window centered by s point, now μ _sreach maximum, maximal value is relevant to the pixel value of target, simultaneously center of gravity G in window _swith window center O _sdistance reduces; When apart from d (G _s, O _s) degree of confidence ρ when ≈ 0 _sreach maximum, simultaneously ρ _sfor local maximum, now window region in region of interest is region of interest.

Here we are called target area gray level image the region of interest obtaining.

(3) use maximum entropy to cut apart to target area gray level image, must arrive target area and cut apart image;

Use maximum entropy to cut apart to target area gray level image, must arrive target area and cut apart image;

If image is divided into two parts of target and background by threshold value Th, the entropy of objective definition and background is respectively:

Background entropy ${\mathrm{Entropy}}_B=-\underset{i=0}{\overset{\mathrm{Th}}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{P_t}\ln \frac{p\left(i\right)}{P_t}$

Target entropy ${\mathrm{Entropy}}_O=-\underset{i=\mathrm{Th}+1}{\overset{m}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{H_t}\ln \frac{p\left(i\right)}{H_t}$

Wherein, P (i) represents that corresponding grey scale level is the probability size of i, the maximum gray scale of m presentation video, P _tand H _trepresent respectively background and target in image pixel grey scale distribute probability and.

Calculate the maximal value of target entropy and background entropy sum as threshold value

$\mathrm{Th}=\underset{\mathrm{Th}}{\arg \max }({\mathrm{Entropy}}_B+{\mathrm{Entropy}}_O)$

If the entropy of certain gray level is greater than Th, think that this gray level is target; If the entropy of certain gray level is less than Th, think that this gray level is background.Can arrive like this target area and cut apart image O (x, y), the image after cutting apart as shown in Figure 4.

(4) utilize radar two dimension echo data and target area gray level image, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix;

Utilize radar two dimension echo data and target area gray level image, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix;

In radar two dimension echo data, scattering center can define 2 class extremal features points: two-dimentional extreme vertex and one dimension extreme point.Definition extreme point:

$p_i=\left\{\begin{array}{cc}1,& \mathrm{if}\min (a_i-a_j)>0\&ForAll;a_j\&Element;U\left(a_i\right)\\ 0,& \mathrm{else}\end{array}\right.$

Wherein U (a _i) represent with a _icentered by local neighborhood (do not comprise a _ipoint).Due to extreme point, in orientation, upwards dynamic range is very large, and therefore the present invention considers orientation one dimension extreme point upwards.

Adopt first order difference method to extract to extreme point defined above.To the point (x, y) in two-dimentional echoed signal, x represent distance to, y represent orientation to, note radar two dimension echo floating point values is f (x, y), definition:

$G(x,y)=\left\{\begin{array}{cc}255,& \mathrm{if}(f(x,y)-f(x,y+1)>0,f(x,y)-f(x,y-1)>0)\\ 0,& \mathrm{else}\end{array}\right.$

Utilizing the G (x, y) that above formula calculates is target area peak point information matrix, as shown in Figure 3.

(5) image is cut apart in target area and target area peak point information matrix carries out information fusion, peak point number K in the target area of image is cut apart in statistics target area, counts out as effective peak.

Image O (x, y) is cut apart to and target area peak point information matrix G (x, y) carries out information fusion in target area, obtain the image R (x, y) after merging.

R(x,y)＝G(x,y)·O(x,y)/255

Image R (x, y) after fusion is 0/255 bianry image, and the pixel that in image, gray level is 255 is target area and cuts apart peak point in objective area in image, adds up these peak point numbers K

$K=C_{\mathrm{ount}}\left\{(x_i,y_i)\right|\left.\begin{array}{c}(x_i,y_i)\&Element;R(x,y)\\ p(x_i,y_i)=255\end{array}\right\}$

Wherein, p (x _i, y _i) be (x in the image R (x, y) after merging _i, y _i) the gray-scale value size located.

(6) peak point in the peak point information matrix of target area is arranged by size, chosen a front K peak point as target effective peak point, binaryzation obtains target effective peak point image.

Peak point in the peak point information matrix of target area, by intensity sequence, is chosen to the effective peak point characteristic information of front K maximum peak point as target.The effective peak point characteristic information of binary conversion treatment target, by this K extreme point place pixel assignment gray level 255, rest of pixels point assignment is 0, obtains target effective peak point image, as shown in Figure 5.

(7) extract the axial feature of target in target effective peak point image, exclusive segment false-alarm point disturbs, and determines target location, as shown in Figure 6.

Before in target effective peak point image, in k candidate target point, may sneak into noise spot or false-alarm point, need to carry out axis projection candidate's impact point is screened.Axis projection has utilized the axial length information of target, and the spatial relation of candidate target point is retrained.

This method adopts the data fitting method of least square method here, and the optimal function that it finds data by the quadratic sum of minimum error is mated.To target pixel points set V, estimate the parameter k of straight line y=kx+b, b, makes residual sum of squares (RSS) minimum, wherein (x _i, y _i) ∈ V.

Axially the algorithm of feature constraint is as follows:

1) in target effective peak point image to representing that k candidate target point utilizes least square fitting, search out a straight line that makes residual sum of squares (RSS) minimum, the axial straight line of target, identical with the diagonal of target;

2) using the length of target, width, Diagonal Dimension as constraint, taking the target rectangle of target length, wide constraint as window, allow the diagonal line of window along the axial rectilinear movement of target, find on the position on axial straight line and make maximum the window's position of contained peak point number in window, the corresponding position of window is target location;

3) the peak point information flag outside window is false-alarm point, and in window, contained peak point information has been determined the position of target.

(8) utilize target location and target energy center of gravity, determine target critical point.

By position and the dimension information of target, can obtain the peak point information in the position of target, calculate the energy barycenter of these peak points as the key point of target, as shown in Figure 7.

According to the characteristic of radar imagery, radar echo intensity shows the information of strong scattering point, and in objective definition of the present invention region, grey scale centre of gravity positions as the key point of target.Grey scale centre of gravity method is " energy " as this point by the gray-scale value of each pixel in region, and the center of gravity formula in its required region is as follows:

$\stackrel{\&OverBar;}{X}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}x\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

$\stackrel{\&OverBar;}{Y}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}y\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

F (x, y) is the floating point values that in radar two dimension echo data, (x, y) locates, and Τ is target area, the grey scale centre of gravity of target area, i.e. the energy barycenter of target area.

Those skilled in the art will readily understand; the foregoing is only preferred embodiment of the present invention; not in order to limit the present invention, all any amendments of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.

Claims (10)

1. the recognition methods of forward-looking radar imaging sea-surface target critical point detection, is characterized in that, described method comprises:

(1) original radar two dimension echo data is quantified as to 2-D gray image data;

(2) 2-D gray image step (1) being obtained utilizes the method for based target physical dimension and degree of confidence to carry out region of interest extraction, obtains target area gray level image;

(3) use maximum entropy to cut apart to target area gray level image, must arrive target area and cut apart image;

(4) utilize radar two dimension echo data and target area gray level image, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix;

(5) image is cut apart in target area and target area peak point information matrix carries out information fusion, in statistics fusion results, peak point number K in target area, counts out as effective peak;

(6) peak point in the peak point information matrix of target area is arranged by size, chosen a front K peak point as target effective peak point, target area peak point information matrix two-value is turned to target effective peak point image;

(7) extract the axial feature of target in target effective peak point image, get rid of false-alarm point and disturb, determine target location;

(8) utilize target location and target energy center of gravity, determine target critical point.

2. the method for claim 1, is characterized in that, described step (1) specifically comprises:

(1.1) the floating point values threshold value that selection quantizes, establishing floating point values upper threshold is L _max, under threshold value, be limited to L _min, wherein:

$L_{\min }=\mathrm{Totalpix}*\frac{\min T}{\mathrm{Totalpix}}$

L _max＝N*(TLength*Margin) ²+L _min，if L _max＜Totalpix

L _max＝Totalpix,if L _max＞Totalpix

N is the target maximum number that may contain in background, Totalpix is the number of pixels of original image, TLength is the pixel count of target length in image, and Margin is surplus to ensure that target can complete demonstration, and minT is that the minimum of target two dimension echo data may floating point values;

(1.2) selective value is L respectively _maxfloating point values and value for L _minfloating point values as threshold value Level255 and Level0; To each data point in original radar two dimension echo data, if being greater than Level255, floating point values gives gray-scale value 255, be less than Level0 and give gray-scale value 0, to being worth at L _maxwith L _minbetween floating point values carry out linear interpolation, determine its gray-scale value, the formula of linear interpolation is as follows:

$g(x,y)=\left\{\begin{array}{cc}0,& \mathrm{if}(f(x,y)<\mathrm{Level}0)\\ \frac{f(x,y)-\mathrm{Level}0}{\mathrm{Level}255-\mathrm{Level}0}*255,& \mathrm{if}(\mathrm{Level}0\&le;f(x,y)\&le;\mathrm{Level}255)\\ 255,& \mathrm{if}(f(x,y)>\mathrm{Level}255)\end{array}\right.$

Wherein f (x, y) is the floating-point numerical value that data point (x, y) is located radar two dimension echo data, and g (x, y) is gray-scale value corresponding to this point (x, y) after linear interpolation.

3. method as claimed in claim 1 or 2, is characterized in that, described step (2) specifically comprises:

(2.1), first according to target sizes and imaging resolution, using the interval under the state of the normal course of target as constraint, determine region of interest window length and width; Investigate the region of interest window centered by s point, in window, the statistic of pixel value has average μ _swith center of gravity G _s:

${\mathrm{\&mu;}}_s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}g(x_i,y_i)$

$\stackrel{\&OverBar;}{x}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}x\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$\stackrel{\&OverBar;}{y}=\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}y\&CenterDot;g(x,y)/\underset{(x,y)\&Element;\mathrm{\&Omega;}}{\mathrm{\&Sigma;}}g(x,y)$

$G_s=(\stackrel{\&OverBar;}{x},\stackrel{\&OverBar;}{y})$

Wherein, n is the number of pixel in the window of region of interest, and g (x, y) is the gray-scale value that in the window of region of interest, (x, y) locates pixel, and Ω is the contained region of region of interest window;

(2.2) ask for the degree of confidence μ of the region of interest window centered by s point _s, and center of gravity G in window _swith window center coordinate O _sdistance d (G _s, O _s), wherein:

ρ _s＝μ _s/[d(G _s,O _s)+1]

d (G _s, O _s) be center of gravity G in window _swith window center coordinate O _sbetween Euclidean distance, wherein x, y is the row and column of presentation video respectively;

(2.3) in 2-D gray image, determine region of interest window according to following principle, as target area gray level image:

If be background in the region of interest window centered by s point, due to background pixel value difference not quite and all lower, now apart from d (G _s, O _s) ≈ 0, degree of confidence ρ _s≈ μ _ssuitable with background pixel value average, now Bu Shi region of interest, region of interest window region;

If a part is that a target part is background, now μ in window in the region of interest window centered by s point _sincrease, but center of gravity G in while window _sbe partial to the region of high pixel value, d (G _s, O _s) also increase, until region of interest window comprises whole target, be defined as region of interest;

If comprised whole target in the region of interest window centered by s point, now μ _sreach maximum, maximal value is relevant to the pixel value of target, simultaneously center of gravity G in window _swith window center O _sdistance reduces; When apart from d (G _s, O _s) degree of confidence ρ when ≈ 0 _sreach maximum, simultaneously ρ _sfor local maximum, now window region in region of interest is region of interest.

4. the method as described in claims 1 to 3 any one, is characterized in that, described step (3) is specially:

(3.1) be identified for segmentation threshold Th that target area gray level image is cut apart, segmentation threshold Th is target entropy Entropy _owith background entropy Entropy _bthe maximal value of sum, wherein:

Background entropy ${\mathrm{Entropy}}_B=-\underset{i=0}{\overset{\mathrm{Th}}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{P_t}\ln \frac{p\left(i\right)}{P_t}$

Target entropy ${\mathrm{Entropy}}_O=-\underset{i=\mathrm{Th}+1}{\overset{m}{\mathrm{\&Sigma;}}}\frac{p\left(i\right)}{H_t}\ln \frac{p\left(i\right)}{H_t}$

P (i) represents that corresponding grey scale level is the probability size of i, the maximum gray scale of m presentation video, P _tand H _trepresent respectively background and target in image pixel grey scale distribute probability and;

(3.2) according to segmentation threshold Th, target area gray level image is cut apart, must be arrived target area and cut apart image O (x, y), be specially:

If the entropy of certain gray level is greater than Th, think that this gray level is target; If the entropy of certain gray level is less than Th, think that this gray level is background, cuts apart image O (x, y) thereby must arrive target area.

5. the method as described in claim 1 to 4 any one, is characterized in that, described step (4) is specially:

According to radar two dimension echo data and target area gray level image, adopt first order difference method, the peak point information in the radar two dimension echo data of extraction target area, obtains target area peak point information matrix G (x, y), wherein:

$G(x,y)=\left\{\begin{array}{cc}255,& \mathrm{if}(f(x,y)-f(x,y+1)>0,f(x,y)-f(x,y-1)>0)\\ 0,& \mathrm{else}\end{array}\right.$

(x, y) represents the point on radar two dimension echo data, x represent distance to, y represent orientation to, remember that its radar two dimension echo floating point values is f (x, y).

6. the method as described in claim 1 to 5 any one, is characterized in that, described step (5) is specially:

Image O (x, y) is cut apart to and target area peak point information matrix G (x, y) carries out information fusion in target area, obtain the image R (x, y) after merging;

R(x,y)＝G(x,y)·O(x,y)/255

Image R (x, y) after fusion is 0/255 bianry image, and the pixel that in image, gray level is 255 is target area and cuts apart peak point in objective area in image, adds up these peak point numbers K,

$K=C_{\mathrm{ount}}\left\{(x_i,y_i)\right|\left.\begin{array}{c}(x_i,y_i)\&Element;R(x,y)\\ p(x_i,y_i)=255\end{array}\right\}$

Wherein, p (x _i, y _i) be (x in the image R (x, y) after merging _i, y _i) the gray-scale value size located.

7. the method as described in claim 1 to 6 any one, is characterized in that, described step (6) is specially:

(6.1) peak point in the peak point information matrix of target area is sorted by size, choose the effective peak point characteristic information of front K maximum peak point as target;

(6.2) the effective peak point characteristic information of target is carried out to binary conversion treatment, by this K extreme point place pixel assignment gray level 255, rest of pixels point assignment is 0, obtains target effective peak point image.

8. the method as described in claim 1 to 7 any one, is characterized in that, described step (7) specifically comprises:

(7.1) in target effective peak point image to representing that k candidate target point utilizes least square fitting, search out a straight line that makes residual sum of squares (RSS) minimum, i.e. the axial straight line of target, this straight line is identical with the diagonal of target;

(7.2) using the length of target, width, Diagonal Dimension as constraint, taking the target rectangle of target length, wide constraint as window, allow the diagonal line of window along the axial rectilinear movement of target, find on the position on axial straight line and make maximum the window's position of contained peak point number in window, the corresponding position of window is target location;

(7.3) the peak point information flag outside window is false-alarm point, and in window, contained peak point information has been determined the position of target.

9. method as claimed in claim 8, is characterized in that, the axial straight line that calculates target in described step (7.1) is specially:

To target pixel points set V, estimate the parameter k of straight line y=kx+b, b, makes residual sum of squares (RSS) minimum, wherein (x _i, y _i) ∈ V.

10. the method as described in claim 1 to 9 any one, is characterized in that, described step (8) is specially:

Ask for the grey scale centre of gravity of target area set it as the energy barycenter of target area, be the key point of target, wherein to ask for formula as follows for the grey scale centre of gravity of target area:

$\stackrel{\&OverBar;}{X}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}x\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

$\stackrel{\&OverBar;}{Y}=\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}y\&CenterDot;f(x,y)/\underset{(x,y)\&Element;T}{\mathrm{\&Sigma;}}f(x,y)$

F (x, y) is the floating point values that in radar two dimension echo data, (x, y) locates, and Τ is target area.