CN101944233A - Method for quickly extracting airport target in high-resolution remote sensing image - Google Patents

Abstract

The invention discloses a method for quickly extracting an airport target in a high-resolution remote sensing image, which is used for solving the technical problem of poor quality of the airport target extracted by a traditional identifying method of the airport target. The technical scheme comprises the steps of: pretreating the image by fuzzy enhancement; then extracting the edges of the image by a fuzzy edge detection method based on pixel gradient and standard difference; screening out the edges of the image to reserve long and straight lines only; detecting parallel long and straight lines as runway characteristics by using Hough change; and with the point with the highest gray within the range of the field of a characteristic point 8 as a seed point, carrying out regional growth for extracting the airport target. The quality of the extracted airport target is enhanced.

Description

The method of rapid extraction airport target in the high-resolution remote sensing images

Technical field

The present invention relates in a kind of remote sensing images to extract the method for rapid extraction airport target in the method, particularly high-resolution remote sensing images of airport target.

Background technology

Existing airport target recognition methods is divided into two classes: the airport target that a class is based on image segmentation detects and extracts.And depend on the quality of image segmentation result based on the method for image segmentation, often the robustness of detection algorithm is not strong for the high-resolution remote sensing figure that contains complicated earth surface information.The another kind of airport target detection method that is based on rim detection.Edge detection method has better robustness than image partition method, yet the edge strength threshold value is difficult to choosing then too, omission may occur or cross the inspection phenomenon if selection is improper, this just may have a strong impact on next airport target detection and the quality of extracting.

Summary of the invention

In order to overcome the ropy deficiency of airport target that existing airport target recognition methods is extracted, the invention provides the method for rapid extraction airport target in a kind of high-resolution remote sensing images, the fuzzy enhancing of this method utilization carries out pre-service to image, utilizes the fuzzy edge detection method based on pixel gradient and standard deviation to extract the image border then; The image border screened only stay long straight line, and utilize the parallel long straight line of Hough change-detection as the runway feature; With in unique point 8 territories the point of high gray scale carry out region growing and extract airport target as seed points, can improve the quality of the airport target that extracts.

The technical solution adopted for the present invention to solve the technical problems: the method for rapid extraction airport target in a kind of high-resolution remote sensing images is characterized in comprising the steps:

(a) by the T conversion

μ _mn＝T(X _mn)＝1-(X _max-X _mn)/D (1)

With pending image I mage, m * n pixel, L level gray scale; From gray space G={G _MnBe mapped as with it corresponding broad sense degree of membership space P={ μ _Mn; In the formula, X _MaxMaximum gradation value among the presentation video G; D is a constant, gets D=2 * (X _Max-X _Min)/3; X _MinMinimum gradation value among the presentation video G;

By generalized fuzzy operator GFO

${\mathrm{\&mu;}}_{\mathrm{mn}}^{\&prime;}=\mathrm{GFO}\left[{\mathrm{\&mu;}}_{\mathrm{mn}}\right]=\left\{\begin{array}{cc}-{[-((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})]}^f& (-r\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}<0)\\ {\left[((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})\right]}^f& (0\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}\&le;r)\end{array}\right.---\left(2\right)$

Processed pixels degree of membership μ ' _Mn=GFO[μ _Mn]; In the formula, r and f are constants, and its scope is 0＜r≤1 and f＞0; When-r≤μ _Mn＜0 o'clock, μ ' _Mn≤ μ _MnAs 0≤μ _Mnμ during≤r _Mn≤ μ ' _Mn

Pass through T ^-1Inverse transformation X _Mn=X _Max-D * (1-μ _Mn) degree of membership P ' is carried out inverse transformation, degree of membership space P ' is mapped as gray space image G ' after the enhancing, finish image is strengthened pre-service;

(b) utilize the sobel operator that the Grad and the gradient direction of each pixel of image are calculated earlier, the pixel gradient value is projected in the scope of [1-100],, utilize formula as an input value of fuzzy system

$\mathrm{Im}{\mathrm{SD}}_{P_5}=\sqrt{\frac{1}{9}\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}{(P_i-E\left(P_i\right))}^2}---\left(3\right)$

Calculate the variance yields of each pixel; In the formula, Pi (i=1,2 ..., 9) remarked pixel point and around eight field points; Its value is also projected in the scope of [1-100], as another input value of fuzzy system; Define four constant threshold: a1, a2, c1, c2; Gradient and variance are divided into high H, middle M, low L three classes; On dutyly then belong to low value class SDL/GDL in scope [0, c1], SD represents standard deviation, and GD represents gradient; If in [a1, c2] scope, then belong to intermediate value class SDM/GDM; In [a2,100], belong to high value class SDH/GDH; The output of fuzzy system is the probability that a pixel belongs to frontier point, is divided three classes too, is respectively EL from high to low, EM, EH;

By formula

P _final＝∑(C _j×P _Edge(j)) (4)

Calculating the border degree of membership of image every bit, if value then is a frontier point greater than threshold value, otherwise is background dot; Obtain the edge of image testing result; In the formula, j is EL, EM, one of EH three classes, C _jRepresent the border degree of membership of j class, P _Edge(j) the current point of representative is under the jurisdiction of the probability of j class;

(c) utilize gradient phase information and line segment length to reject short or curvilinear lengths, and utilize the Hough variation that remaining outline line is carried out parallel lines and detect, retrieve the characteristic image S of parallel straight line as airfield runway;

(d) the characteristic image S of traversal airfield runway judges by the visit sign picture whether it is accessed, if its corresponding point that indicates is that 0 representative is not accessed, is that 1 representative is accessed; If accessed then get back to step (a) and continue to search, be designated as Sp otherwise preserve this point coordinate;

Be central point with Sp among the picture G ' behind fuzzy the enhancing, travel through its eight field, find the highest point of gray scale as planting child node; Preserve its coordinate, be designated as SeedP;

With a SeedP is that seed points begins region growing, and up to all satisfactory points, all growing finishes, and the point identification that all-access is crossed all is made as 1; Judging whether image S travels through finishes; If then the picture as a result after the output area growth finishes; Then return step (a) if not and continue to search unique point.

The invention has the beneficial effects as follows:, utilize fuzzy edge detection method to extract the image border then based on pixel gradient and standard deviation owing to utilize fuzzy the enhancing that image is carried out pre-service; The image border screened only stay long straight line, and utilize the parallel long straight line of Hough change-detection as the runway feature; With in unique point 8 territories the point of high gray scale carry out region growing and extract airport target as seed points, improved the quality of the airport target that extracts.

Below in conjunction with the drawings and specific embodiments the present invention is elaborated.

Description of drawings

Fig. 1 is the method flow diagram of rapid extraction airport target in the high-resolution remote sensing images of the present invention.

Fig. 2 is that the method for rapid extraction airport target in the high-resolution remote sensing images of the present invention is with eight field template figure.

Fig. 3 is the method fuzzy rule figure of rapid extraction airport target in the high-resolution remote sensing images of the present invention.

Embodiment

With reference to Fig. 1～3.

1. fuzzy enhancing pre-service.

More accurate for the extraction that makes target, at first image is strengthened pre-service.Its concrete steps are as follows:

(1) by the T conversion

μ _mn＝T(X _mn)＝1-(X _max-X _mn)/D (1)

With pending image I mage, m * n pixel, L level gray scale; From gray space G={G _MnBe mapped as with it corresponding broad sense degree of membership space P={ μ _Mn; In the formula, X _MaxMaximum gradation value among the presentation video G; D is a constant, generally gets D=2 * (X _Max-X _Min)/3; X _MinMinimum gradation value among the presentation video G.

(2) by generalized fuzzy operator GFO

${\mathrm{\&mu;}}_{\mathrm{mn}}^{\&prime;}=\mathrm{GFO}\left[{\mathrm{\&mu;}}_{\mathrm{mn}}\right]=\left\{\begin{array}{cc}-{[-((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})]}^f& (-r\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}<0)\\ {\left[((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})\right]}^f& (0\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}\&le;r)\end{array}\right.---\left(2\right)$

Processed pixels degree of membership μ ' _Mn=GFO[μ _Mn], on this process nature a kind of process of fuzzy enhancing.In the formula, r and f are constants, and its scope is 0＜r≤1 and f＞0.Getting the f value here is 0.5, and the r value is 1.When-r≤μ _Mn＜0 o'clock, μ ' _Mn≤ μ _MnAs 0≤μ _Mnμ during≤r _Mn≤ μ ' _MnObviously the generalized fuzzy operator is by reduction-r≤μ _Mnμ in＜0 zone _MnValue and increase by 0≤μ _Mnμ in the≤r zone _MnValue, played the effect that strengthens contrast between two zones.

(3) pass through T ^-1Inverse transformation X _Mn=X _Max-D * (1-μ _Mn) degree of membership P ' is carried out inverse transformation, degree of membership space P ' is mapped as gray space image G ' after the enhancing, thereby finishes the process of whole fuzzy enhancing.

2. Image Edge-Detection.

Utilize the sobel operator that the Grad and the gradient direction of each pixel of image are calculated earlier, the pixel gradient value is projected in the scope of [1-100], as an input value of fuzzy system.Utilize formula

$\mathrm{Im}{\mathrm{SD}}_{P_5}=\sqrt{\frac{1}{9}\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}{(P_i-E\left(P_i\right))}^2}---\left(3\right)$

Calculate the variance yields of each pixel.In the formula, Pi (i=1,2 ..., 9) be expressed as pixel and around eight field points.Its value is also projected in the scope of [1-100], as another input value of fuzzy system.Then define four constant threshold: a1, a2, c1, c2.Here, c1=12, a1=37, a2=62, c2=87.Gradient and variance are divided into high H, middle M, low L three classes.On dutyly then belong to low value class SDL/GDL in scope [0, c1], SD represents standard deviation, and GD represents gradient.If in [a1, c2] scope, then belong to intermediate value class SDM/GDM.In [a2,100], belong to high value class SDH/GDH.The output of fuzzy system is the probability that a pixel belongs to frontier point, is divided three classes too, is respectively EL from high to low, EM, EH.The degree of membership that this three class is under the jurisdiction of frontier point is respectively 0.25,0.5 and 0.75.

By formula

P _final＝∑(C _j×P _Edge(j)) (4)

Calculate the border degree of membership of image every bit.In the formula, j is EL, EM, one of EH three classes, C _jRepresent the border degree of membership of j class, P _Edge(j) the current point of representative is under the jurisdiction of the probability of j class.

After obtaining the border degree of membership of this point, it is judged.If value then is a frontier point greater than threshold value, otherwise is background dot.In general judgment threshold should get 0.5.Obtain the edge of image testing result thus.

Illustrate: if the standard difference SD of a pixel is 75, Grad GD is 30, according to four threshold value a1, a2, c1, c2.(12,37,62,87) can calculate its SD be under the jurisdiction of low in the degree of membership of Senior Three class be respectively μ (SD _L)=0,

The degree of membership that GD is under the jurisdiction of low middle Senior Three class is respectively

μ (GD _H)=0.Then according to fuzzy rule:

P _Edge(EL)＝μ(GD _L)×μ(SD _M)＝0.28×0.48＝0.13

P _Edge(EM)＝μ(GD _M)×μ(SD _M)+μ(GD _L)×μ(SD _H)＝0.49

P _Edge(EH)＝μ(GD _M)×μ(SD _H)＝0.72×0.52＝0.37

P _final＝∑(C _j×P _Edge(j))＝C _EL×P _Edge(EL)+C _EM×P _Edge(EM)+C _EH×P _Edge(EH)＝0.56

P _Final＞0.5 this pixel is judged as frontier point.

3. airport target feature extraction.

The edge wheel profile that utilizes the gradient phase information of marginal point eliminate to disturb: from up to down seek marginal point in edge image, i.e. white point is after finding white point Pe, be made as current point, seek its eight field, if find white point Pe ', difference to 2 gradient phase informations judges, if | OG _Pe-OG _{Pe '}|≤threshold, threshold=0.3 can judge that then it is roughly the point on same the straight line here.If Pe ' then looks for full straight line for current point continues to seek up to can not find satisfactory point, add up and write down the contained pixel number of this straight line afterwards.Repeating above step all marginal point traversals in edge image finishes.What remain so basically all is the edge contour that is similar to straight line, gets appropriate threshold again to remove too short straight line, is only contained the edge image of long straight line.

To carrying out the Hough conversion through the garbled edge image of long straight line to seek the feature of parallel long straight line as the airport.The Hough conversion is the polar equation that adopts straight line: r=xsin θ+ycos θ.(r, θ) middle corresponding curve can meet at same point (r to all points at parameter space on same the straight line ₀, θ ₀).Then a local maximum in parameter space has just been represented the straight line in the image space.If the θ value of two straight lines is identical two straight line parallels are described then.The pixel contained detected parallel lines is designated white, just obtained the image S of airfield runway feature.

4. extraction airport target.

Extracted after the airfield runway feature, also needed airport target is extracted.Since material, reasons such as smoothness, and the gray scale in most of airfield runways zone can be than land area height.Can extract airport target according to this feature.Arthmetic statement is as follows:

(1) traversal characteristic image S searches the airport feature point.

(2) find unique point after, judge by the visit sign picture whether it accessed, if its corresponding point that indicates is that 0 representative is not accessed, be that 1 representative is accessed.If accessed then rebound step 1 continues to search, be designated as Sp otherwise preserve this point coordinate.

(3) be central point with Sp among the picture G ' behind fuzzy the enhancing, travel through its eight field, find the highest point of gray scale as planting child node.Preserve its coordinate, be designated as SeedP.

With a SeedP is that seed points begins region growing, and up to all satisfactory points, all growing finishes.The difference that is gray-scale value and seed points absolute value is less than certain threshold value, and threshold value gets 20 here.And the point identification that all-access is crossed all is made as 1.Judging whether image S travels through finishes.If, the picture as a result after then output area is grown, algorithm finishes.Then return step 1 if not and continue to search unique point.

Claims (1)

1. the method for rapid extraction airport target in the high-resolution remote sensing images is characterized in that comprising the steps:

(a) by the T conversion

μ _mn＝T(X _mn)＝1-(X _max-X _mn)/D (1)

With pending image I mage, m * n pixel, L level gray scale; From gray space G={G _MnBe mapped as with it corresponding broad sense degree of membership space P={ μ _Mn; In the formula, X _MaxMaximum gradation value among the presentation video G; D is a constant, gets D=2 * (X _Max-X _Min)/3; X _MinMinimum gradation value among the presentation video G;

By generalized fuzzy operator GFO

${\mathrm{\&mu;}}_{\mathrm{mn}}^{\&prime;}=\mathrm{GFO}\left[{\mathrm{\&mu;}}_{\mathrm{mn}}\right]=\left\{\begin{array}{cc}-{[-((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})]}^f& (-r\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}<0)\\ {\left[((r^{1/f-1}+r)\&CenterDot;{\mathrm{\&mu;}}_{\mathrm{mn}}-{{\mathrm{\&mu;}}^2}_{\mathrm{mn}})\right]}^f& (0\&le;{\mathrm{\&mu;}}_{\mathrm{mn}}\&le;r)\end{array}\right.---\left(2\right)$

Processed pixels degree of membership μ ' _Mn=GFO[μ _Mn]; In the formula, r and f are constants, and its scope is 0＜r≤1 and f＞0; When-r≤μ _Mn＜0 o'clock, μ ' _Mn≤ μ _MnAs 0≤μ _Mnμ during≤r _Mn≤ μ ' _Mn

Pass through T ^-1Inverse transformation X _Mn=X _Max-D * (1-μ _Mn) degree of membership P ' is carried out inverse transformation, degree of membership space P ' is mapped as gray space image G ' after the enhancing, finish image is strengthened pre-service;

(b) utilize the sobel operator that the Grad and the gradient direction of each pixel of image are calculated earlier, the pixel gradient value is projected in the scope of [1-100],, utilize formula as an input value of fuzzy system

$\mathrm{Im}{\mathrm{SD}}_{P_5}=\sqrt{\frac{1}{9}\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}{(P_i-E\left(P_i\right))}^2}---\left(3\right)$

Calculate the variance yields of each pixel; In the formula, Pi (i=1,2 ..., 9) remarked pixel point and around eight field points; Its value is also projected in the scope of [1-100], as another input value of fuzzy system; Define four constant threshold: a1, a2, c1, c2; Gradient and variance are divided into high H, middle M, low L three classes; On dutyly then belong to low value class SDL/GDL in scope [0, c1], SD represents standard deviation, and GD represents gradient; If in [a1, c2] scope, then belong to intermediate value class SDM/GDM; In [a2,100], belong to high value class SDH/GDH; The output of fuzzy system is the probability that a pixel belongs to frontier point, is divided three classes too, is respectively EL from high to low, EM, EH;

By formula

P _final＝∑(C _j×P _Edge(j)) (4)

Calculating the border degree of membership of image every bit, if value then is a frontier point greater than threshold value, otherwise is background dot; Obtain the edge of image testing result; In the formula, j is EL, EM, one of EH three classes, C _jRepresent the border degree of membership of j class, P _Edge(j) the current point of representative is under the jurisdiction of the probability of j class;

(c) utilize gradient phase information and line segment length to reject short or curvilinear lengths, and utilize the Hough variation that remaining outline line is carried out parallel lines and detect, retrieve the characteristic image S of parallel straight line as airfield runway;

(d) the characteristic image S of traversal airfield runway judges by the visit sign picture whether it is accessed, if its corresponding point that indicates is that 0 representative is not accessed, is that 1 representative is accessed; If accessed then get back to step (a) and continue to search, be designated as Sp otherwise preserve this point coordinate;

Be central point with Sp among the picture G ' behind fuzzy the enhancing, travel through its eight field, find the highest point of gray scale as planting child node; Preserve its coordinate, be designated as SeedP;

With a SeedP is that seed points begins region growing, and up to all satisfactory points, all growing finishes, and the point identification that all-access is crossed all is made as 1; Judging whether image S travels through finishes; If then the picture as a result after the output area growth finishes; Then return step (a) if not and continue to search unique point.

Images