CN101976347A - Method for recognizing overwater bridge in remote sensing image on basis of Mean Shift segmentation - Google Patents

Abstract

The invention discloses a method for recognizing an overwater bridge in a remote sensing image on basis of Mean Shift segmentation, which is used for solving the technical problem of low recognition rate of the existing overwater bridge target recognition method. The technical scheme is as follows: carrying out image segmentation by utilizing a Mean Shift method and color features, extracting a river region, and determining a river domain by utilizing similarity; combining expansion and corrosion to communicate the river region, extracting a bridge region according to the features of a bridge; refining to obtain a communicated river skeleton line, and finding the intersection point of the skeleton line and the bridge region; and finally recognizing the bridge by utilizing the shape and textural features of the bridge, thereby improving the recognition rate of recognizing an overwater bridge target.

Description

On-water bridge recognition methods in the remote sensing images of cutting apart based on Mean Shift

Technical field

The present invention relates to a kind of remote sensing images recognition methods, on-water bridge recognition methods in particularly a kind of remote sensing images of cutting apart based on Mean Shift.

Background technology

Document " based on the medium-and-large-sized on-water bridge Target Recognition of the aerial image of knowledge; Wuhan University of Technology's journal; 2005; Vol.29 (2); p230-233 " discloses the medium-and-large-sized on-water bridge target identification method of a kind of aerial image, and this method has proposed a kind of bridge target identification method based on knowledge at the high-altitude image of taking photo by plane.According to the strong relativity between waters and territory, land, the bridge territory, aerial image is carried out two-value cut apart, come in waters and other Region Segmentation, according to the waters bridge is carried out Primary Location.Use the seed points growth method accurately to mark the bridge territory then, think that the width of bridge is very little for length, so the axial straight line that obtains through the HOUGH conversion process obtains the width of bridge according to the pixel on both sides on the axis.But, because bridge shared ratio in image is less, the background complexity, the gray scale contrast ratio is less, is difficult under the data-driven by image segmentation, extracts target signature and then judges recognition objective.And think that the primary feature of bridge is to have two parallel long straight lines, but in the actual photographed, owing to the visual angle of sensor, the reason of capture distance are not to have parallel long straight line.So the described method of document has limitation when handling remote sensing images, discrimination is lower.

Summary of the invention

In order to overcome the low deficiency of existing on-water bridge target identification method discrimination, the invention provides on-water bridge recognition methods in a kind of remote sensing images of cutting apart based on Mean Shift.This method adopts Mean Shift method, utilizes color characteristics to carry out image segmentation, extracts river region, and utilizes similarity to determine river valley.Be communicated with river region in conjunction with expanding, corroding,, extract the bridge zone according to the feature of bridge.The skeleton line in the river that obtains being communicated with by refinement, the intersection point of seeking with the bridge zone finds candidate's bridge, utilizes the shape textural characteristics identification bridge of bridge at last, can improve the discrimination of on-water bridge Target Recognition.

The technical solution adopted for the present invention to solve the technical problems: on-water bridge recognition methods in a kind of remote sensing images of cutting apart based on Mean Shift is characterized in may further comprise the steps:

(a) the remote sensing images format conversion is arrived the LUV space, remote sensing images are carried out Mean Shift cut apart and the zone merging;

The order zone merges original pixel { x in the associating territory, back _iI=1,2..., n; Pixel { z after the filtering in the associating territory _iI=1,2..., n; Cut apart the back remote sensing images in i pixel be labeled as L _i, i=1,2..., n.

Utilize gaussian kernel function k (x) to estimate the characteristic density space, any point in the remote sensing images drifted about with gaussian kernel function k (x):

$k\left(x\right)={\left(2\mathrm{\Π}\right)}^{-d/2}\exp (-\frac{1}{2}{\left|\right|x\left|\right|}^2)---\left(1\right)$

Utilize k (x) to draw and have constringent recursion formula

$y_{i+1}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}---\left(2\right)$

With Mean Shift vector

$m_h\left(x\right)=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}-x=y_{i+1}-y---\left(3\right)$

Carry out the iteration convolution, up to satisfying stopping criterion, promptly displacement is less than setting number || m _h(x)-and x ||＜ε or drift number of times reach maximal value; In the formula, h is a bandwidth parameter, and h=(hs, hv), hs is a spatial domain color characteristic bandwidth, hv is a spatial bandwidth, { c _jIt is the negative derivative of the section function of kernel function k (x);

Remote sensing images are carried out Mean Shift filtering and the information all about convergence point all is kept at z _iIn;

In the associating territory, generate cluster { c _jJ=1 ..., m, all in spatial domain distance less than hv and at the z of colourity territory distance less than hs _iCombine;

For any i=1,2..., n makes Li={j|z _i∈ c _j, set Minimum Area M, reject area of space less than M;

(b) similar value between the calculating All Ranges, the vector in each zone are carried out similarity as inner product and are judged;

Perhaps, the three-component that each is regional is converted into gray-scale value, asks each regional regional area variance, carries out similarity and judges; And with minimum regional area variance, as first plot of river;

Perhaps, calculate mean square deviation judgement similarity between each zone of residue and the first river region;

Choose successively satisfy above similarity standard the zone as river region;

(c) remote sensing images are carried out binary conversion treatment, utilize the potential river region of connected component labeling method mark, and remove the noise river region; River region is carried out expansive working, be communicated with the river; Do poorly with largest connected waters with the former binaryzation river region that is not communicated with then, utilizing between two river region is the characteristics in bridge territory, extracts the bridge zone, and river axis is extracted in refinement;

(d) seek river valley center line and the intersection point that is not communicated with the river image, and drop on the point in bridge zone, extract candidate's bridge; Extract the bridge outline line, utilize the shape textural characteristics to discern bridge.

Described bandwidth parameter h=(hs, hv) in, spatial domain color characteristic bandwidth hs optimum value is 9, spatial bandwidth hv optimum value is 8.5.

Described Minimum Area M is 500.

The invention has the beneficial effects as follows: owing to adopt Mean Shift method, utilize color characteristics to carry out image segmentation, extract river region, and utilize similarity to determine river valley.Be communicated with river region in conjunction with expanding, corroding,, extract the bridge zone according to the feature of bridge.The skeleton line in the river that obtains being communicated with by refinement, the intersection point of seeking with the bridge zone finds candidate's bridge, utilizes the shape textural characteristics identification bridge of bridge at last, has improved the discrimination of on-water bridge Target Recognition.

Below in conjunction with embodiment the present invention is elaborated.

Embodiment

1. at first the remote sensing images format conversion of the colour of input is arrived the LUV space, remote sensing images are carried out MeanShift cut apart.

1) carry out Mean Shift filtering:

Utilize gaussian kernel function k (x) to estimate the characteristic density space, any point in the remote sensing images (not repeating) is drifted about with gaussian kernel function k (x).

$k\left(x\right)={\left(2\mathrm{\Π}\right)}^{-d/2}\exp (-\frac{1}{2}{\left|\right|x\left|\right|}^2)---\left(1\right)$

Utilize k (x) to draw then and have constringent recursion formula (2) and Mean Shift vector (3), carry out the iteration convolution, up to satisfying stopping criterion, displacement is less than setting number || m _h(x)-and x||＜ε or drift number of times reach maximal value, and Mean Shift vector is normalized probability density gradient, and constantly the gradient direction along probability density moves, and always points to the maximum direction that probability density increases.

$y_{i+1}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}---\left(2\right)$

$m_h\left(x\right)=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}-x=y_{i+1}-y---\left(3\right)$

In the formula, h is a bandwidth parameter, and h=(hs, hv), hs is a spatial domain color characteristic bandwidth, hv is a spatial bandwidth, selects empirical value in the application for use.{ c _jIt is the negative derivative of the section function of kernel function k (x).

Mean Shift partitioning algorithm has merged image space territory and color gamut information, the resolution when bandwidth parameter is determining filtering.These two parameters have significant effects for Mean Shift filter effect.Experiment finds that hs is provided with 9, and it is 8.5 proper that hv is set to.

2) zone merges:

Realization merges the zone of remote sensing images, and zone adjacent and that the pixel value difference is little is merged, and obtains a bigger zone.Concrete steps are: make { x _iI=1,2..., n and { z _iI=1,2..., n represent pixel after the original and filtering in the associating territory respectively, the label L of i pixel in the image of back is cut apart in order _i, i=1,2..., n represents.

A) remote sensing images are carried out Mean Shift filtering and the information all about convergence point all is kept at z _iIn;

B) in the associating territory, generate cluster { c _jJ=1 ..., m, all in spatial domain distance less than hv and at the z of colourity territory distance less than hs _iCombine;

C) for any i=1,2..., n makes Li={j|z _i∈ c _j, set Minimum Area M, present embodiment is got M=500, rejects those less than the M area of space.

2. the LUV three-component is rotated back into rgb space again.Sorted by size in each zone of Mean Shift the deletion zonule.All Ranges is carried out similarity judges, extract river region:

1) similar value between the calculating All Ranges, the vector in each zone are carried out similarity as inner product and are judged.

2) each is regional three-component is converted into gray-scale value, asks each regional regional area variance, carries out similarity and judges.And with minimum regional area variance, as first plot of river.

3) mean square deviation of calculating between each zone of residue and the first river region is judged similarity.

It is final that what adopt is that the similarity of comprehensively carrying out of above-mentioned three kinds of methods judges that judgment criterion is an empirical value.Choose satisfy above similarity standard the zone as river region.

3. image is carried out binary conversion treatment, utilize the potential river region of connected component labeling method mark, and remove the noise river region.River region is carried out expansive working, be communicated with the river.Do poorly with largest connected waters with the former binaryzation river region that is not communicated with then, utilizing between two river region is the characteristics in bridge territory, extracts the bridge zone.And river axis is extracted in refinement.

4. seek the river valley center line and intersect and drop on the intersection point that is equipped with the candidate region, obtain the bridge candidate region with the former binaryzation river region that is not communicated with.Determine the bridge number according to the sum of river region, and extract the bridge outline line, utilize the shape textural characteristics to discern bridge.

Claims (3)

1. on-water bridge recognition methods in the remote sensing images of cutting apart based on Mean Shift is characterized in that comprising and has write step:

(a) the remote sensing images format conversion is arrived the LUV space, remote sensing images are carried out Mean Shift cut apart and the zone merging;

The order zone merges original pixel { x in the associating territory, back _iI=1,2..., n; Pixel { z after the filtering in the associating territory _iI=1,2..., n; Cut apart the back remote sensing images in i pixel be labeled as L _i, i=1,2..., n.

Utilize gaussian kernel function k (x) to estimate the characteristic density space, any point in the remote sensing images drifted about with gaussian kernel function k (x):

$k\left(x\right)={\left(2\mathrm{\Π}\right)}^{-d/2}\exp (-\frac{1}{2}{\left|\right|x\left|\right|}^2)---\left(1\right)$

Utilize k (x) to draw and have constringent recursion formula

$y_{i+1}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}---\left(2\right)$

With Mean Shift vector

$m_h\left(x\right)=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_{i}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}g\left({\left|\right|\frac{{x-x}_i}{h}\left|\right|}^2\right)}-x=y_{i+1}-y---\left(3\right)$

Carry out the iteration convolution, up to satisfying stopping criterion, promptly displacement is less than setting number || m _h(x)-x||＜ε or the drift number of times reach maximal value; In the formula, h is a bandwidth parameter, and h=(hs, hv), hs is a spatial domain color characteristic bandwidth, hv is a spatial bandwidth, { c _jIt is the negative derivative of the section function of kernel function k (x);

Remote sensing images are carried out Mean Shift filtering and the information all about convergence point all is kept at z _iIn;

In the associating territory, generate cluster { c _iJ=1 ..., m, all in spatial domain distance less than hv and at the z of colourity territory distance less than hs _iCombine;

For any i=1,2..., n makes Li={j|z _i∈ c _j, set Minimum Area M, reject area of space less than M;

(b) similar value between the calculating All Ranges, the vector in each zone are carried out similarity as inner product and are judged;

Perhaps, the three-component that each is regional is converted into gray-scale value, asks each regional regional area variance, carries out similarity and judges; And with minimum regional area variance, as first plot of river;

Perhaps, the mean square deviation of calculating between each zone of residue and the first river region is judged similarity;

Choose satisfy above similarity standard the zone as river region;

(c) remote sensing images are carried out binary conversion treatment, utilize the potential river region of connected component labeling method mark, and remove the noise river region; River region is carried out expansive working, be communicated with the river; Do poorly with largest connected waters with the former binaryzation river region that is not communicated with then, utilizing between two river region is the characteristics in bridge territory, extracts the bridge zone, and river axis is extracted in refinement;

(d) seek river valley center line and the intersection point that is not communicated with the river image, and drop on the point in bridge zone, extract candidate's bridge; Extract the bridge outline line, utilize the shape textural characteristics to discern bridge.

2. method according to claim 1 is characterized in that: described bandwidth parameter h=(hs, hv) in, spatial domain color characteristic bandwidth hs optimum value is 9, spatial bandwidth hv optimum value is 8.5.

3. method according to claim 1 is characterized in that: described Minimum Area M is 500.