CN102270295B - SAR (synthetic aperture radar) image rapid bridge detection method - Google Patents

Abstract

The invention discloses an SAR (synthetic aperture radar) image rapid bridge detection method for mainly solving the problem that the bridge detection speed is slow in the prior art. The method comprises the following steps: (1) selecting Daubechies5 wavelet to perform the wavelet decomposition on an input SAR image; (2) extracting water contour information in a low-frequency sub-image to determine a potential bridge region in the low-frequency sub-image; (3) mapping the potential bridge region in the low-frequency sub-image to a coordinate system of an original image by using a coordinate mapping formula; (4) performing a region connection on the potential bridge region in the original image; (5) extracting bridge characteristics from the potential bridge region in the original image; (6) judging whether the target in the potential bridge region is a bridge according to the obtained bridge characteristics; and (7) computing a bridge direction of the potential bridge region with the bridge. Compared with the similar method, under the condition of guaranteeing the bridge correct detection rate, the method has the advantages that the computation load is reduced, the bridge detectionspeed is improved, and the real-time requirement in the practical application is satisfied.

Description

The quick bridge detection method of SAR image

Technical field

The invention belongs to the SAR technical field of image processing, specifically a kind of SAR image bridge target method for quick based on wavelet decomposition.

Background technology

The bridge target is positioned at turnpike road and intersection, river mostly as important transport hub, is closely connected with various mobile or fixed military targets, and the bridge detection of therefore studying in the SAR image has great importance.

The SAR image bridge detection method that proposes mainly contains at present:

People such as WU have proposed the method that a kind of Pun of utilization entropy detects bridge in " Bridge recognition of median-resolution SAR images using Pun histogram entropy; Chinese Optics Letters; 2009; 7 (7); 572-575 ", this method at first water system is cut apart extraction water body profile, extracts the Pun entropy feature of bridge then in the potential bridge zone of determining, and utilize this feature to carry out the differentiation of bridge, remove false target.This method is only utilized the Pun feature of bridge and is not considered the shape facility of bridge, and some false targets that reach the entropy threshold value are detected as bridge; Because factors such as radar incident direction, bridge structure can cause some bridge gray-scale value obviously to reduce, can not reach the entropy threshold value and by omission.

People such as Jiang Yongmei have proposed the bridge detection method with SAR image, infrared image, optical imagery fusion in " towards the automatic multi-source remote sensing image co-registration model and method that detects of bridge target; electronics and information journal; 2006; 28 (10): 1794-1797 ", the characteristics that comprehensive utilization multi-source image information has complementation realize the bridge detection.But this method need be obtained identical SAR image, infrared image and the optical imagery of same place three width of cloth resolution respectively, obtaining of image data source, is difficult to meet the demands.

People such as Chen Xin have proposed the multiple dimensioned SAR bridge object detection method of a kind of based target in " the automatic bridge detection method of intermediate-resolution SAR image; computer engineering; 2008; 34 (9): 195-197 ", under the low resolution rank, extract the potential bridge impact point in river and location, again positioning result is mapped in the original medium resolution image data, obtains the bridge testing result.This method uses 10 times to extract and obtain low resolution SAR images, and still there is the SAR speckle noise in the image that obtains, and extraction can make a large amount of useful informations of missing image.

Preceding two kinds of methods all are that at first selected threshold is carried out image and cut apart, and extract river region, and this method operand of cutting apart based on the original graph waters is big, and the speed that bridge detects is slow, is difficult to be applied to in the demanding occasion of real-time; Though last a kind of method is utilized the multiple dimensioned operand that reduced, lost the useful information in the image, reduced the precision that bridge detects, in addition, this method also is difficult to play the good restraining effect to the speckle noise in the SAR image.

Summary of the invention

In order to overcome the existing slow problem of detection method detection speed, the present invention proposes a kind of detection method of bridge fast, under the situation that guarantees the bridge accuracy of detection, to shorten working time, improve the speed that bridge detects, satisfy in the practical application requirement to real-time.

Technical scheme of the present invention is: earlier original image is carried out wavelet decomposition, obtain size and be the low frequency subgraph picture of former Fig. 1/4, in gained low frequency subgraph picture, extract potential bridge zone then, this zone is mapped in the former figure coordinate again, the shape, the gray feature that fully utilize bridge at last detect differentiation to the target in the potential zone, remove false target.Its specific implementation step comprises as follows:

1) selects for use the Daubechies5 small echo that the SAR image of input is carried out wavelet decomposition, obtain the low frequency subgraph picture of original image;

2) extract the water body profile information in the low frequency subgraph picture, determine potential bridge zone, this zone is a rectangular area, and the coordinates table on four summits is shown (x _Min1, y _Min1), (x _Min1, y _Max1), (x _Max1, y _Min1), (x _Max1, y _Max1);

3) the potential bridge zone in the low frequency subgraph picture is utilized coordinate mapping formula be mapped in the original image coordinate system, obtain the potential bridge zone in the original image, this zone still is the rectangular area, and the coordinates table on its four summits is shown (x _Min2, y _Min2), (x _Min2, y _Max2), (x _Max2, y _Min2), (x _Max2, y _Max2), the potential bridge zone in the original image is numbered, be numbered 1,2...N, N is total number in potential bridge zone,

Coordinate mapping formula is as follows:

$\left\{\begin{array}{c}{x}_{\min 2}=2(x_{\min 1}-f)\\ y_{\min 2}=2(y_{\min 1}-g)\\ x_{\max 2}=2(x_{\max 1}-f)\\ x_{\max 2}=2(y_{\max 1}-g)\end{array}\right.$

Wherein, f, g represent respectively to be mapped to the x axle of original image coordinate time, y axial coordinate side-play amount, f=(2m-m from low frequency subgraph as coordinate _x)/2, g=(2n-n _x)/2, m * n is low frequency subgraph picture size, m _x* n _xBe the original image size;

4) to the potential bridge zone of the N after the numbering, be divided into according to the combined method in the mathematics

Group, two zone numbers in every group are respectively a and b, calculate the entropy H in these two potential bridge zones respectively _a, H _b

5) ask for the slope k of dividing two zones in each group of back _a, k _bAnd between them apart from s;

6) if following decision condition is satisfied in these two zones simultaneously, then these two zones are connected:

H _a＞T _eAnd H _b＞T _e, T wherein _eFor the entropy threshold value, be set to 5.2 according to experiment;

| k _a-k _b|＜T _k, T wherein _kBe slope threshold value, be set to 0.5;

S＜T _s, T wherein _sBe distance threshold, be set to 5;

7) with after the connection of zone, extract the bridge target in each potential bridge zone, according to the potential bridge of the rectangle zone among the former figure of bridge target update that extracts, the rectangular top point coordinate after the renewal is (x _Min3, y _Min3), (x _Min3, y _Max3), (x _Max3, y _Min3), (x _Max3, y _Max3), extract the edge length L of bridge in each potential bridge zone respectively _eLength L with bridge _b, and should potential bridge zone be that the boundary is divided into four little rectangular areas with the rectangular centre, note rectangular area, upper left side is A, and the rectangular area, upper right side is B, and the rectangular area, lower left is C, and the rectangular area, lower right is D, extracts each regional gray average I respectively _A, I _B, I _CAnd I _D

8) according to the bridge feature that obtains, judge whether the target in the potential bridge zone is bridge:

If the length L of bridge _bThe edge length L of ∈ [8,480] and bridge _e＜2.9L _b, then think this potential bridge region memory at bridge, otherwise think and do not have bridge;

9) to there being the potential bridge zone of bridge, utilize each regional gray average I _A, I _B, I _CAnd I _DCalculate the direction of bridge: if I _B＞I _AAnd I _C＞I _D, then bridge is the BC diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction; If I _A＞I _BAnd I _D＞I _C, then bridge is the AD diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction.

The present invention compared with prior art has the following advantages:

1. detection speed is fast.The present invention is owing to extract potential bridge zone in the low frequency subgraph picture, in original image, only detect the target in the potential bridge zone, therefore dwindle the hunting zone of bridge target greatly, the operand of detection algorithm has been reduced to original 1/4, improved the efficient that detects.

2. need not to adopt the speckle noise in the complicated processing inhibition image.Because wavelet decomposition itself has good effect to image noise reduction, so after the original SAR image among the present invention carried out wavelet decomposition, the low frequency subgraph picture that obtains was compared more level and smooth with original image, speckle noise is inhibited.

Theoretical analysis and simulation result show that the present invention compared with prior art under the situation that guarantees the bridge verification and measurement ratio, has improved detection speed.

Description of drawings

Fig. 1 is implementing procedure figure of the present invention;

Fig. 2 is original SAR image;

Fig. 3 is that the present invention carries out the low frequency subgraph picture that obtains after the wavelet decomposition to original image;

Fig. 4 is that the present invention tests the potential bridge zone in the low frequency subgraph picture that obtains;

Fig. 5 is that the present invention tests the potential bridge of the former figure zone that obtains;

Fig. 6 is that the present invention tests the final bridge testing result figure that obtains.

Embodiment

With reference to Fig. 1, under the specific implementation block diagram of the present invention:

Step 1 selects for use the Daubechies5 small echo that the SAR image of input is carried out wavelet decomposition, obtains the low frequency subgraph picture of original image.

Step 2 is extracted the water body profile information in the low frequency subgraph picture, determine potential bridge zone.

(2a) the draw histogram of low frequency subgraph picture is chosen the threshold value T that first trough point corresponding gray scale value is cut apart as the waters in the histogram _r, according to this threshold value low frequency subgraph is looked like to carry out the waters and cut apart and obtain the water body zone, and mark is carried out in the water body zone that obtains, be designated as 1,2...t successively, wherein, t is the number in water body zone;

(2b) to each water body zone that obtains, adopt its edge of Canny operator extraction, and to the t behind the mark potential bridge zone, be divided into according to the combined method in the mathematics

Group is asked for every group of bee-line d between two edges, if d is greater than setting threshold L _r, judging between these two coastal waters does not have bridge, continue to calculate in other group the bee-line between two coastal waters, if d is less than setting threshold L _r, execution in step (2c), L _rBe set to 10 according to experiment;

(2c) for bee-line less than L _rTwo coastal waters because the secondary lobe influence of bridge when having imaging, all satisfy distance less than the pixel of 1.5d in these two edges so need find, and obtain the minimum horizontal ordinate x in these pixels _Min1, minimum ordinate y _Min1, maximum horizontal ordinate x _Max1, maximum ordinate y _Max1, with (x _Min1, y _Min1), (x _Min1, y _Max1), (x _Max1, y _Min1), (x _Max1, y _Max1) being summit structure rectangle, this rectangle is the potential bridge zone in the low frequency subgraph picture.

Step 3 utilizes coordinate mapping formula to be mapped in the original image coordinate system in the potential bridge zone in the low frequency subgraph picture, obtains the potential bridge zone in the original image, and this zone still is the rectangular area, and the coordinates table on its four summits is shown (x _Min2, y _Min2), (x _Min2, y _Max2), (x _Max2, y _Min2), (x _Max2, y _Max2), the potential bridge zone in the original image is numbered, be numbered 1,2...N, N is total number in potential bridge zone,

Coordinate mapping formula is as follows:

$\left\{\begin{array}{c}{x}_{\min 2}=2(x_{\min 1}-f)\\ y_{\min 2}=2(y_{\min 1}-g)\\ x_{\max 2}=2(x_{\max 1}-f)\\ x_{\max 2}=2(y_{\max 1}-g)\end{array}\right.---\left(1\right)$

Wherein, f, g represent respectively to be mapped to the x axle of original image coordinate time, y axial coordinate side-play amount, f=(2m-m from low frequency subgraph as coordinate _x)/2, g=(2n-n _x)/2, m * n is low frequency subgraph picture size, m _x* n _xBe the original image size.

Step 4 to the potential bridge zone of the N after the numbering, is divided into according to the combined method in the mathematics

Group, two zone numbers in every group are respectively a and b, calculate the entropy H in these two potential bridge zones respectively according to following formula _a, H _b:

$H_a=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\ln p_i$

$H_b=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{q}_i\ln q_i$

Wherein, i is the gray level of image pixel, p _iFor gray level among the potential bridge zone a is the probability of the pixel appearance of i, q _iFor gray level among the potential bridge zone b is the probability of the pixel appearance of i.

Step 5 is asked for the slope k of dividing two zones in each group of back _a, k _bAnd between them apart from s:

k _a＝(y _max2(a)-y _min2(a))/(x _max2(a)-x _min2(a))

k _b＝(y _max2(b)-y _min2(b))/(x _max2(b)-x _min2(b))

$s=\sqrt{{(x_{\min 2}\left(b\right)-x_{\max 2}\left(a\right))}^2+{(y_{\min 2}\left(b\right)-y_{\max 2}\left(a\right))}^2}$

Wherein, (x _Min2, y _Min2) be the coordinate in the upper left corner, rectangle potential bridge zone in the original image, (x _Max2, y _Max2) be the coordinate in the lower right corner, rectangle potential bridge zone.

Step 6 according to two decision conditions that satisfy in the zone in each group, determines whether these two zones are connected: if H is satisfied in these two zones simultaneously _a＞T _e﹠amp; H _b＞T _e﹠amp; | k _a-k _b|＜T _k﹠amp; S＜T _s, then these two zones are connected; Otherwise, execution in step 7, wherein, T _eFor the entropy threshold value, be set to 5.2, T according to experiment _kBe slope threshold value, be set to 0.5, T _sBe distance threshold, be set to 5.

Step 7 after the connection of zone, is extracted the bridge target in each potential bridge zone, according to the coordinate figure in the potential bridge of rectangle zone among the former figure of bridge target update that extracts, calculates the edge length L of bridge in each the potential bridge zone after upgrading respectively _eLength L with bridge _b, and extract the gray feature in this potential bridge zone simultaneously.

(7a) with the gray-scale value in the potential bridge zone greater than bridge segmentation threshold T _bPixel value be made as 1, all the other are made as 0, T _bBe set to T according to experiment _r+ 0.28, then pixel value is that 1 part is the bridge target, obtains the minimum horizontal ordinate x of these pixel values _Min3, minimum ordinate y _Min3, maximum horizontal ordinate x _Max3With maximum ordinate y _Max3, with (x _Min3, y _Min3), (x _Min3, y _Max3), (x _Max3, y _Min3), (x _Max3, y _Max3) be summit structure rectangle, remember that this rectangle is the potential bridge zone after upgrading;

(7b) adopt the profile track algorithm to calculate the edge length L of bridge _e, its concrete steps are as follows:

(7b1) according to from left to right, the potential bridge zone of from top to bottom direction search after upgrading, with first marginal point of obtaining as initial trace point (X _q, Y _q), the start edge length of note bridge is 0;

(7b2) centered by initial trace point, adopt 8 direction chain codes, direction search according to since 0 to 7, be 1 marginal point up to searching pixel value, if this marginal point is 1,3,5,7 directions with respect to initial trace point, then the bridge edge length is updated to initial bridge edge length and adds 1, if this marginal point is 0,2,4,6 directions with respect to initial trace point, then the bridge edge length is updated to initial bridge edge length and adds

And with the initial trace point of this marginal point as next step search, with the initial bridge edge length of the bridge edge length after upgrading as next step search;

(7b3) to new initial trace point and initial bridge edge length, repeating step (7b) is up to gained new initial trace point and (X _q, Y _q) overlap, to follow the tracks of and finish, the bridge edge length of this moment is L _e

(7c) calculate the length L of bridge according to following formula _b:

$L_b=\sqrt{{(x_{\max 3}-x_{\min 3})}^2+{(y_{\max 3}-y_{\min 3})}^2}$

Wherein, (x _Min3, y _Min3) be the coordinate in the upper left corner, rectangle potential bridge zone after upgrading, (x _Max3, y _Max3) be the coordinate in the lower right corner, rectangle potential bridge zone;

(7d) the potential bridge zone after will upgrading is that the boundary is divided into four little rectangular areas with the rectangular centre, note rectangular area, upper left side is A, and the rectangular area, upper right side is B, and the rectangular area, lower left is C, the rectangular area, lower right is D, extracts each regional gray average I respectively _A, I _B, I _CAnd I _D, and with this as this regional gray feature.

Step 8, according to the bridge feature that obtains, judge whether the target in the potential bridge zone is bridge:

If bridge length L _b∈ [8,480] and bridge edge length L _e＜2.9L _b, then think this potential bridge region memory at bridge, otherwise think and do not have bridge.

Step 9 to there being the potential bridge zone of bridge, is utilized each regional gray average I _A, I _B, I _CAnd I _DCalculate the direction of bridge: if I _B＞I _AAnd I _C＞I _D, then bridge is the BC diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction; If I _A＞I _BAnd I _D＞I _C, then bridge is the AD diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction.

Effect of the present invention can specify by emulation experiment:

1. experiment condition

Test used microcomputer and be configured to Intel (R) Core (TM) 2Duo CPU, 2.53GHz, the 1.98GB internal memory, Windows XP operating system, programming platform is Matlab7.0.1.The view data that adopts in the experiment is that area, Washington resolution is the SAR image of 1m, and the image size is 402 * 514, downloads from the Sandia of U.S. National Laboratory website.

2. experiment content

This experiment is divided into determines that in the low frequency subgraph picture potential bridge zone, coordinate mapping and bridge detect three parts:

The original SAR image that experiment is used at first carries out wavelet decomposition to Fig. 2 as shown in Figure 2, obtains size and is the low frequency subgraph picture of former Fig. 1/4, as shown in Figure 3, in Fig. 3, carry out the waters then and cut apart and extract the water body profile information, obtain the potential bridge zone in the low frequency subgraph picture, as shown in Figure 4.

After in the low frequency subgraph picture, obtaining potential bridge zone, according to coordinate mapping formula, the potential bridge zone in the low frequency subgraph picture is mapped in the former figure coordinate system, obtains the potential bridge zone in the original image, as shown in Figure 5.

After obtaining the potential bridge zone in the original image, at first carrying out the zone connects, extract bridge clarification of objective in each zone then, and according to these features the target in this zone is detected differentiation, remove false target, after whole experiment is 4.3s working time, obtain final detection result as shown in Figure 6, wherein the center of bridge, length and direction are as shown in table 1:

Table Bridge 1 beam testing result

As can be seen from Table 1,4 bridges to be detected in the original image all are detected successfully, have obtained position, length and the directional information of bridge simultaneously, have realized the location to bridge.

Claims (2)

1. the quick bridge detection method of SAR image comprises the steps:

1) selects for use the Daubechies5 small echo that the SAR image of input is carried out wavelet decomposition, obtain the low frequency subgraph picture of original image;

2) extract the water body profile information in the low frequency subgraph picture, determine potential bridge zone, this zone is a rectangular area, and the coordinates table on four summits is shown (x _Min1, y _Min1), (x _Min1, y _Max1), (x _Max1, y _Min1), (x _Max1, y _Max1);

3) the potential bridge zone in the low frequency subgraph picture is utilized coordinate mapping formula be mapped in the original image coordinate system, obtain the potential bridge zone in the original image, this zone still is the rectangular area, and the coordinates table on its four summits is shown (x _Min2, y _Min2), (x _Min2, y _Max2), (x _Max2, y _Min2), (x _Max2, y _Max2), the potential bridge zone in the original image is numbered, be numbered 1,2 ... N, N are total number in potential bridge zone,

Coordinate mapping formula is as follows:

$\left\{\begin{array}{c}{x}_{\min 2}=2(x_{\min 1}-f)\\ y_{\min 2}=2(y_{\min 1}-g)\\ x_{\max 2}=2(x_{\max 1}-f)\\ y_{\max 2}=2(y_{\max 1}-g)\end{array}\right.$

Wherein, f, g represent respectively to be mapped to the x axle of original image coordinate time, y axial coordinate side-play amount, f=(2m-m from low frequency subgraph as coordinate _x)/2, g=(2n-n _x)/2, m * n is low frequency subgraph picture size, m _x* n _xBe the original image size;

4) to the potential bridge zone of the N after the numbering, be divided into according to the combined method in the mathematics Group, two zone numbers in every group are respectively a and b, calculate the entropy H in these two potential bridge zones respectively _a, H _b

5) ask for the slope k of dividing two zones in each group of back _a, k _bAnd between them apart from s;

6) if following decision condition is satisfied in these two zones simultaneously, then these two zones are connected:

H _a＞T _eAnd H _b＞T _e, T wherein _eFor the entropy threshold value, be set to 5.2 according to experiment;

| k _a-k _b|＜T _k, T wherein _kBe slope threshold value, be set to 0.5;

S＜T _s, T wherein _sBe distance threshold, be set to 5;

7) with after the connection of zone, extract the bridge target in each potential bridge zone, according to the potential bridge of the rectangle zone among the former figure of bridge target update that extracts, the rectangular top point coordinate after the renewal is (x _Min3, y _Min3), (x _Min3, y _Max3), (x _Max3, y _Min3), (x _Max3, y _Max3), the edge length L of bridge in the potential bridge zone after extracting each respectively and upgrading _eLength L with bridge _b, and should potential bridge zone be that the boundary is divided into four little rectangular areas with the rectangular centre, note rectangular area, upper left side is A, and the rectangular area, upper right side is B, and the rectangular area, lower left is C, and the rectangular area, lower right is D, extracts each regional gray average I respectively _A, I _B, I _CAnd I _D

The edge length L of bridge in the potential bridge zone after each renewal of described extraction _eLength L with bridge _b, carry out as follows:

(7a) according to from left to right, the potential bridge zone of from top to bottom direction search after upgrading, first marginal point that obtains is initial trace point (X _q, Y _q), the initial bridge edge length of note bridge is 0;

(7b) centered by initial trace point, adopt 8 direction chain codes, direction search according to since 0 to 7, be 1 marginal point up to searching pixel value, if this marginal point is 1,3,5,7 directions with respect to initial trace point, then the bridge edge length is updated to initial bridge edge length and adds 1, if this marginal point is 0,2,4,6 directions with respect to initial trace point, then the bridge edge length is updated to initial bridge edge length and adds

, and with the initial trace point of this marginal point as next step search, with the initial bridge edge length of the bridge edge length after upgrading as next step search;

(7c) to new initial trace point and initial bridge edge length, repeating step (7b) is up to the new initial trace point that obtains and (X _q, Y _q) overlap, to follow the tracks of and finish, the bridge edge length of this moment is L _e

(7d) calculate the length L of bridge according to following formula _b:

$L_b=\sqrt{{(x_{\max 3}-x_{\min 3})}^2+{(y_{\max 3}-y_{\min 3})}^2}$

Wherein, (x _Min3, y _Min3) be the coordinate in the upper left corner, rectangle potential bridge zone after upgrading, (x _Max3, y _Max3) be the coordinate in the lower right corner, rectangle potential bridge zone;

8) according to the bridge feature that obtains, judge whether the target in the potential bridge zone after upgrading is bridge:

If the length L of bridge _bThe edge length L of ∈ [8,480] and bridge _e＜2.9L _b, then think this potential bridge region memory at bridge, otherwise think and do not have bridge;

9) to there being the potential bridge zone of bridge, utilize each regional gray average I _A, I _B, I _CAnd I _DCalculate the direction of bridge: if I _B＞I _AAnd I _C＞I _D, then bridge is the BC diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction; If I _A＞I _BAnd I _D＞I _C, then bridge is the AD diagonal, calculates the angle of this direction and horizontal direction, namely gets the bridge direction.

2. the quick bridge detection method of SAR image according to claim 1 is characterized in that, the entropy H in the potential bridge of the described calculating of step 4) zone _a, H _b, undertaken by following formula:

$H_a=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\ln p_i$

$H_b=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{q}_i\ln q_i$

Wherein, i is the gray level of image pixel, p _iFor gray level among the potential bridge zone a is the probability of the pixel appearance of i, q _iFor gray level among the potential bridge zone b is the probability of the pixel appearance of i.

Images