CN103714339B - SAR image road damaging information extracting method based on vector data - Google Patents
SAR image road damaging information extracting method based on vector data Download PDFInfo
- Publication number
- CN103714339B CN103714339B CN201310750917.9A CN201310750917A CN103714339B CN 103714339 B CN103714339 B CN 103714339B CN 201310750917 A CN201310750917 A CN 201310750917A CN 103714339 B CN103714339 B CN 103714339B
- Authority
- CN
- China
- Prior art keywords
- road
- vector data
- sar image
- area
- width
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
Abstract
Provided is an SAR image road damaging information extracting method based on vector data. According to the range of SAR images after a disaster, vector data of a corresponding area are obtained; the vector data are projected to a coordinate system of the SAR images and then are registered on the SAR images; a suspected road damaging area of the SAR images is extracted, wherein a road detecting operator is used for line detecting, road width information and the road vector data are subjected to shape level set segmentation, and the road damaging area is obtained by fusion; and a Bayes network model is established to carry out further judging on the suspected road damaging area, and road damaging information is extracted. According to the method, the vector data are used as prior information for helping SAR image road changing detecting, detecting rate is high, a breaking zone can be well extracted through a method with line detecting and a shape level set being combined, the omission factor is low, interference information can be effectively removed through the established Bayes network model, and false alarm in road damaging extracting is lowered.
Description
Technical field
The present invention relates to Remote Sensing Image Processing Technology field, particularly to the high-resolution under a kind of vector data auxiliary
Sar image road damages information extracting method.
Background technology
Road is national economy and military tremulous pulse, all has very important significance on military and civilian.When various disasters
When generation, road life line may be blocked, for example flood, landslide, and the natural disaster such as mud-rock flow all may lead to
The blocking on road, so that sending rescue personnel and greatly being hindered toward disaster area transport rescue material, brings huge to rescue and relief work
Big inconvenience.After natural disaster occurs, because the mode of artificial field exploring is the work taking time and effort, remote sensing technology
Extract the particularly important mode of road damage because the feature of " sky eye " makes.
Make a general survey of the research that scholars extract in recent years to remote sensing image road damage, Most scholars are all using multispectral
Optical image to carry out road damage extraction as data source, and the quick obtaining of optical image disaster-stricken after boisterous shadow
Sound larger so as to be limited by very large in rescue and relief work decision-making.Sar, due to its special imaging mechanism, can overcome
Weather and the impact of illumination condition, carry out round-the-clock, round-the-clock, on a large scale observation, therefore occur in disaster to target area
Afterwards, extract road using sar image, to find out road damage region more advantageous.The sar despite many scholar's research
The method that image road extracts, but the road damage extraction for sar image, especially high-resolution sar image then rarely has
People sets foot in.
Content of the invention
For the problems referred to above, the present invention adopts vector data as the damage extraction of auxiliary guiding road it is achieved that high score
The road damage of resolution sar image extracts.
The technical scheme is that a kind of damage information extracting method of the sar image road based on vector data, including
Following steps:
Step 1, according to the scope of the sar image after calamity, obtains the vector data of corresponding region, described vector data includes
Road vectors data；
Step 2, after the vector data of step 1 gained corresponding region is projected to the coordinate system of sar image, by vector number
According to being registrated on sar image；
Step 3, extracts the doubtful road damage area of sar image, including following sub-step,
Step 3.1, in the relief area set up according to road vectors data, enters line detection using Road Detection operator,
Obtain road line segment primitive and road width information, find the position of road fracture according to the testing result of road line segment primitive,
Obtain doubtful road damage area；
Step 3.2, step 3.1 gained road width information is combined with road vectors data, obtains shape level set and divide
The prior shape constraint cut, carries out level set movements using prior shape constraint, obtains the segmentation result of road area and find
The fracture position of road, obtains doubtful road damage area；
Step 3.3, step 3.1 and 3.2 gained doubtful road damage area is merged, obtains final doubtful road damage
Area；
Step 4, sets up the doubtful road damage area that Bayesian network model extracts to step 3 and determines whether,
Extract road damage information.
And, in step 1, download vector data from openstreetmap server.
And, in step 1, described vector data also includes outline of house vector data and waters contour vector data.
And, step 1 includes following sub-step,
Step 1.1,4 angular coordinates of sar image are projected under the coordinate system of vector data, if projection gained longitude and latitude
Degree coordinate is respectively (x1,y1), (x2,y2), (x3,y3), (x4,y4)；
Step 1.2, asks for x1,x2,x3,x4Maximum, minima xmax,xminAnd y1,y2,y3,y4Maximum, minima
ymax,ymin；
Step 1.3, by (xmin,ymax), (xmin,ymin), (xmax,ymax), (xmax,ymin) as 4 angles downloading scope
Point, obtains the vector data of corresponding region.
And, in step 3.1, described Road Detection operator is as follows,
The template of Road Detection operator be width be 2w, length be the rectangle of l, the middle section of template be width be w,
Length is the rectangle of l, and the left and right sides region respectively width of template is w/2, length is the rectangle of l；Using template meter
Calculate response value gap=min(1-avr2/avr1,1-avr2/avr3), if calculating gap < 0, make gap=0.
And, in step 3.1, enter line detection using Road Detection operator, obtain road line segment primitive and road width
Information, implementation is,
If certain width sar image resolution is n rice, make w1=8/n, w2=16/n, w3=24/n, w4=32/n, to road vectors number
According in every section of road vectors set road width w=w successively1,w2,w3,w4, carry out following operation under each value, in this section of road
Road vector both sides is extended, and sets up the relief area that overall width is 4w, and mobile Road Detection operator is detected in relief area,
Record this line segment when response value gap is more than predetermined threshold value；The move mode of mobile Road Detection operator is to swear from this section of road
Amount starting point starts, and buffering is divided into several minibuffer areas, the template of each minibuffer section length and mobile Road Detection operator
Length is consistent；In each minibuffer area, from road vectors centrally along perpendicular to road vectors direction toward two side shiftings, width
Often it is separated by w/4 to calculate once；Record in the template accordingly moving Road Detection operator under the result of calculation that each exceedes threshold value
Heart line segment, and calculate the line segment sum detecting in relief area；
The line segment sum relatively detecting in relief area under each value, selection detects that the value of most line segments is
The developed width of road.
And, in step 3.2, to every section of road vectors in road vectors data, using this section of road vectors as centrage,
Width according to road is extended to the shape constraining information as road for the long strip type region of a closure, long strip type area toward both sides
The length in domain is the length of certain section of road, and width is the width of road.
And, in step 4, described Bayesian network model includes 6 priori evident information variable a, b, c, d, e, f, 2
Observation g, h and doubtful road damage area actual attribute x, 6 priori evident information variable a, b, c, d, e, f are doubtful road path losses
Ruin the condition of area actual attribute x, doubtful road damage area actual attribute x is 2 observations g, the condition of h；Doubtful by solving
The Posterior probability distribution of the actual attribute in road damage area is as follows, the maximum situation of select probability as final result of determination,
Wherein, p (x/a, b, c, d, e, f) represents that doubtful road damage area belongs to certain situation under the conditions of various evidences
Prior probability, p (g/x), p (h/x) represents the relation between the attribute in doubtful road damage area and observation respectively.
And, described 6 priori evident information variable a, b, c, d, e, f respectively outline of house vector data, landslide is hidden
Suffer from point data, calamity kind and its intensity, dsm data, road vectors data, waters contour vector data；2 observations g, h is respectively
For doubtful road damage area gray scale, doubtful road damage area texture.
The present invention proposes the method that a kind of line detection of improvement relief area and level-set segmentation combine, and extracts sar image
The change of road, and then set up Bayesian network model with reference to the observation of other aucillary documents and fracture zone, to these changes
Region is further judged, rejects false detection, extracts the true damage information of road.
Brief description
Fig. 1 is the general flow chart of the embodiment of the present invention.
Fig. 2 is the window model figure of the Road Detection operator of the embodiment of the present invention.
Fig. 3 is the move mode figure in relief area for the Road Detection template of the embodiment of the present invention.
Fig. 4 is the combination fracture zone observation of the embodiment of the present invention and the Bayesian network model figure of aucillary document.
Specific embodiment
The present invention provides the sar image road damage extracting method under a kind of geographical vector data auxiliary.Main with
The damage of high-resolution sar image road is extracted as research contents, using geographical vector data as prior information, based on changing
The relief area line detection entered and the integrated method of level-set segmentation find doubtful road damage area.Consider further in sar image
Folded cover, the factor such as the interference of coherent spot and road background complexity of itself, the present invention takes Bayes posterior probability model
Depth analysis are carried out to doubtful road damage area, and then extracts road damage.
Describe technical solution of the present invention below in conjunction with drawings and Examples in detail.
Openstreetmap (abbreviation osm) is a Internet map cooperation plan increased income, and user can be free from net
The various vector datas of upper download (include road vectors data, house and water body contour vector data etc.), and its real-time property is strong
And high precision, the damage of prior information relief road can be work perfectly well as and extract.Embodiment adopts disclosed geography information arrow
Amount data openstreetmap auxiliary, provides high-resolution sar image road damage extracting method.
Technical solution of the present invention can realize automatic running using computer software technology.As shown in figure 1, the technology of embodiment
Protocol procedures comprise the following steps:
The acquisition of step one vector data.The up-to-date osm road of corresponding scope can automatically be obtained according to sar image capturing range
Vector data, outline of house vector data, waters contour vector data, the every kind of vector data in sar image capturing range may divide
Do not include multistage vector line segment, every section of vector line segment is made up of a series of point.Wherein road vectors data is used as number before calamity
According to source, the change information of assisted extraction road；Latter two vector data can by as aucillary document in order to change information to be
No for damage judgement for further analysis.When being embodied as, can be according to the scope of post-disaster high-resolution sar image, from net
The upper vector data automatically downloading corresponding region.
Embodiment uses openstreetmap api (xapi), selects to download region, builds a range boundary frame, then
Build a download address, download vector initial data from openstreetmap server.The determination mode downloading scope is:
(1) 4 angular coordinates of sar image are projected under the coordinate system of openstreetmap vector data, if projection
Gained latitude and longitude coordinates are respectively (x1,y1), (x2,y2), (x3,y3), (x4,y4).
(2) ask for x1,x2,x3,x4Maximum, minima xmax,xminAnd y1,y2,y3,y4Maximum, minima ymax,ymin.
(3) by (xmin,ymax), (xmin,ymin), (xmax,ymax), (xmax,ymin) as 4 angle points downloading scope, obtain
Take the vector data of corresponding region.
In practical operation situation, there is certain deviation due on vector data and sar image coordinate, therefore download
Scope actual may be slightly larger than scope determined by said method.
After the vector data of corresponding region is projected to the coordinate system of sar image by step 2, vector data is registrated to
On sar image.
After embodiment obtains openstreetmap vector data and is projected into sar coordinate systems in image, sar image
Grid deviation is there may be and vector data between.When being embodied as, can be using sar image as benchmark, with reference in prior art
The registering mode of same place or line of the same name corrects openstreetmap vector data, including to step one gained road vectors number
According to, outline of house vector data, waters contour vector data all carry out registration.
If the obvious cross point of big measure feature and flex point are existed on image, such as in road vectors data and respective image
Road there is cross point or flex point, then take select same place registering mode more suitable, in vector data and sar
Select corresponding cross point or flex point as same place to rear on image respectively, you can using polynomial correction in prior art
Method calculates the coefficient of conversion.Then to each point on each section of vector line segment, according to polynomial equation with counted
The conversion coefficient calculating calculates the position after correction.For the sake of improving efficiency, man-machine interaction circle when being embodied as, can be passed through
Face (such as touch screen) provides a user with vector data and sar image, and same place can be specified by user and be selected.
Be difficult to situation about seeking in view of some image flex points and cross point, can in the way of using line of the same name it is only necessary to
The line segment being located on the same line is selected on vector data and sar image, does not need strictly to mate between line segment.For letter
Change the operation of user, with the design software method of operation can be, on vector data, user only needs to reconnaissance, and software can be certainly
Dynamic travel through all of vector line segment, and find out from the nearest line segment of this point (2 points of straightways being linked to be on vector) as being selected
Line segment.Select the concrete steps suggestion of line segment registration as follows:
(1) several uneven line segments are selected on sar image.In order to ensure the precision of registration, choose between line segment
Angle can not be too little.
(2) corresponding vector line segment is selected on vector data, practising way is, when user's chosen distance corresponds to line segment
Near point, system can travel through all vector line segments automatically, choose comprise this point (the vertical point of this spot projection to line segment positioned at line segment it
Interior) and the minimum vector line segment of distance.
(3) intersection point of line segments is asked for respectively according to the line segment selecting on sar image and vector data.
(4) using the cross point asked in step (3) as same place, using selection same place registration described above
Vector data is registrated on image method.I.e. can be utilized prior art in polynomial correction method calculate conversion be
Number, then to each point on each section of vector, calculates and entangles with the conversion coefficient that calculated according to polynomial equation
Position after just.
Subsequent step is carried out according to the image after registration and vector.
The method of the detection of step 3 joint line and shape level-set segmentation extracts the doubtful road damage area of sar image.
Embodiment implementation is as follows,
(1) under the auxiliary of the openstreetmap road vectors data after step 2 registration, according to road vectors
In the relief area that data is set up, enter line detection using Road Detection operator, the line segment primitive of road can be obtained and have a lot of social connections in road
Degree information.Not will detect that the thought of road axis primitive based on fracture zone, can be tied according to the detection of road line segment primitive
Fruit finds the position of road fracture, obtains a part of doubtful road damage area.
(2) the road width information that line detection obtains is combined by the present invention with road vectors data, as shape level set
The prior shape constraint of segmentation.Carry out level set movements using prior shape constraint, also can get the segmentation result of road area
And find the fracture position of road, also obtain a part of doubtful road damage area.
(3) doubtful road damage area in line testing result and shape level-set segmentation result is merged, to reduce leakage
The occurrence of inspection.
In (1), the method obtaining road width is based on the idea that the middle section width when road detection template
More cater to the developed width of certain section of road, then detect the road line segment primitive obtaining in this section of relief area using this template
Sum will be more.
In (2), the method setting up the constraint of shape level set prior shape is: using each section of road vectors line as center
Line, the width according to road is extended to the long strip type region of a closure toward both sides, and the length in long strip type region is certain section of road
Length, width be road width.
Embodiment, under the auxiliary of openstrertmap road vectors data, is calculated using the Road Detection improving d1 operator
Son detects road-center line primitives and road width information.The mould of the Road Detection operator that the present invention adopts after improving d1 operator
Accompanying drawing 2 is shown in by plate.Wherein template width is 2w, and middle section width is the road width that w(w is selection, is selectable variable)；
L be template length, that is, template be width be 2w, length be l rectangle, in template length be l center line segment can be by template
It is divided into two width to be w, length is the rectangle of l.The operator of embodiment mainly considers central area and left and right region, template
Middle section for width be w, length be the rectangle of l, the left and right sides region of template be respectively width be w/2, length be l's
Rectangle.When being embodied as, l can calculate according to road vectors line segment length, a length of s of such as certain section road vectors line segment, and one
As default l be 2 times of template width w, then this section of road vectors line segment can be divided into n=integer(s/l) integer section, the most at last
The length correction of l is l=s/n, and integer represents and rounds.If its middle section (as black region part in Fig. 2) gray average
For avr2, left and right sides area grayscale average is respectively avr1, avr3.Side using response value gap of this line segment of formwork calculation
Method is:
Gap=min(1-avr2/avr1,1-avr2/avr3)；
If above formula calculates gap < 0, make gap=0, otherwise preserve above formula result of calculation constant.
Road is divided into various ranks according to the difference of its function, and the road width of different stage is different.2 tracks, 4 cars
Road, 6 tracks, the road in 8 tracks are that we are modal, and the impact due to the other footpath of road, in high-resolution sar
The developed width that on image, road presents is also somewhat larger.What the embodiment of the present invention was rough is divided into 8 road according to width
Rice, 16 meters, 24 meters, 32 meters of several ranks it is assumed that certain width sar image resolution be n rice, then road on sar image can
The width selecting is respectively w1=8/n, w2=16/n, w3=24/n, w4=32/n.When being embodied as, can be set according to real road situation
Put rank.
Every section of road vectors in road vectors data are carried out process and obtain Road by the method that embodiment utilizes line detection
Duan Jiyuan and road width method particularly includes:
(1) first suppose that road width w is w1, then corresponding Road Detection operator width is 2w1, the width of middle section is
w1, then it is extended certain section of road vectors both sides, setting up overall width is 4w1Relief area.Movement Road Detection operator (see
Above-mentioned improvement d1 operator and the Road Detection operator that comes) detected in relief area, when operator response value gap is more than default threshold
This line segment is recorded during value (those skilled in the art voluntarily can preset value it is proposed that value is between 0.12 0.2).Detective operators
Move mode be: be 4w in width1Relief area in, Road Detection template, from the beginning of this section of road vectors starting point, will cache
Divide into the segment buffer area of several sections (i.e. n sections), each sectored cells domain is designated as a minibuffer area, length is l.Each
In individual minibuffer area, from road vectors centrally along perpendicular to road vectors direction, toward two side shiftings, width is often separated by w1/ 4 calculating
Once, if road vectors center is designated as 0, by template at 0 toward a side shifting to w1/4、w1/2、3w1/4、w1、5w1/4、3w1/
2、7w1/4、2w1Place calculates respectively, calculates in the same movement of opposite side and respectively.Calculate for 17 times in a minibuffer area
The tailing edge relief area of one-tenth moves down length l of a template, continues to repeat aforesaid operations in the minibuffer area of next part,
Till whole buffer detection finishes.Record and accordingly move Road Detection operator under the result of calculation that each exceedes threshold value
The line segment sum that detects in relief area of template center's line segment calculating.Mobile side in relief area for the Road Detection template
Formula is as shown in Figure 3.
(2) road width is changed into w2, w3, w4Re-establish relief area respectively again to be calculated with Road Detection proposed by the present invention
Son calculates the response value of each position operator, and recording responses value exceedes the line segment of threshold value.
(3) if the middle section width of Road Detection operator more agrees with the developed width of road, then in relief area
Detect satisfactory line segment also more.Therefore comparison operator middle section width is respectively w1, w2, w3, w4Do not sympathize with
The line segment bar number detecting in relief area under condition, selects to detect the reality that the most middle section width of line segment bar number is road
Border width, and specifically detected the primary data of the testing result of most line segments as subsequent treatment, detected with current
The line segment that each going out exceedes threshold value is road primitives.
Line segment primitive and the road of each section of road vectors after above-mentioned road buffering area line detecting step, are obtained
Width information.Due to being incomplete coincidence (see shown in accompanying drawing 2) between Mei Duan minibuffer area, so even being that road does not occur
Fracture, also likely to be present trickle fracture between road primitives.When being embodied as, after detecting road primitives, can be utilized
The information such as the distance between line segment primitive, curvature, Curvature varying, road primitives are carried out marshalling becomes broken line.Additionally, due to
Both sides of the road there may be interference information (with roadway characteristic similar the moon that for example in a row building formed similar with road
Shadow), also marshalling result to be screened therefore after marshalling.The algorithm of screening is to be weaker than real road based on interference information
The thought of road information is carried out: for the broken line after marshalling, if the overlap that two broken lines project in road vectors
Scope reaches certain threshold value (can be set to the 1/3 of shorter broken line), the road line segment primitive hop count being comprised according to broken line, rejects
The broken line of line segment primitive negligible amounts, retains the larger broken line of line segment primitive quantity.After organizing into groups and screening step,
The broken line obtaining is projected on road vectors, wherein there is not the road vectors scope of line segment projection, be then identified as
The region of fracture.
And the road width information obtaining through the detection of relief area line then combines with road vectors data, formed shape water
The prior shape constraint of flat collection.Concrete grammar is to every section of road vectors in road vectors data, using this section of road vectors as
Centrage, the width according to road is extended to the shape constraining information as road for the long strip type region of a closure toward both sides,
The length in long strip type region is the length of certain section of road, and width is the width of road.After obtaining the shape constraining information of road,
Obtain the road area of every section of road vectors using the method segmentation of shape level-set segmentation in prior art.Every section of road is sweared
The road area of amount can carry out segmentation judgement, for example, carry out segmentation according to l, and according to every segment road area institute occupied road picture
Plain number number to judge whether to rupture, (road pixel number shared by for example every segment road area is less than and accordingly presets
It is judged as during threshold value rupturing), thus finding out the region of road fracture.The present invention adds the shape constraining of road target, uses
The prior shape of target is carried out the segmentation of bound level collection and just can effectively be rejected interference, obtains and more accurately extracts result.
In order to reduce missing inspection that may be present, present invention two methods based on line detection and shape level-set segmentation by this
The doubtful road damage area extracting combines, will both results merge (can using ask and by the way of merge).
Step 4 utilizes various auxiliary informations and doubtful road damage area observation, sets up Bayesian network model, to step
The rapid three doubtful road damage areas extracting are determined whether, extract the damage information of road.
In traditional road damage extracting method, simply according to image after calamity before calamity, (image before calamity can be using road arrow
Amount data replaces) extract road change information, the region then changing these is as the damage region of road.But it is actual
On due to the impact of other various interference, road change information that change-detection obtains be not necessarily true damage caused
, therefore present invention employs Bayesian network model and judgement is further analyzed to the fracture zone of detection, have higher
Reliability.
The present invention is divided into following a few class situation the doubtful damage region of road sar image above extracting:
(1) road both sides building collapses after earthquake, leads to road congestion.
(2) the landslide blocked path that earthquake or other reasonses cause.
(3) reason such as heavy rain or seismic barrier lake causes road to be inundated with flood.
(4) what high-lager building and hypsography caused folded covers.
(5) complicated intersection and bridge.
(6) coherent speckle noise and other interference.
Determine that the various aucillary documents of its prior probability and these events exist with the inclusion that above this six events are associated
Observation on image.Related with above three damages first is exactly specific calamity kind, such as house collapse be substantially by
In seismic, without generation earthquake, then it is considered that because the probability that house collapse causes road damage is 0,
Probability in the event of earthquake so house collapse is also relevant with concrete earthquake magnitude.In addition also have a lot of reasons and this six events
Correlation, is illustrated below respectively.
House collapse main with whether occur earthquake and its earthquake magnitude relevant, but the road damage that causes of house collapse then with
House is related to the distance of road.If there is not house this section of road both sides, then this section of region is because house falls
Collapse and cause the probability damaged to be just 0.The distance of therefore house and road is one of the reason causes this kind of damage.
The generation of landslide is also relevant with the geological conditions in this region, if there are landslide point data, then just
Can landslide point data with this at occur landslide cause the probability of road damage to connect.In addition this section of road side mountain region
The gradient etc. be also the probability of road damage can be caused to link together with landslide, if judging to engage in this profession according to dem/dsm
Road both sides does not have hillside, belongs to level land, then be impossible to landslide yet.
Flood inundation on tracks is then relevant with the elevation residing for this section of road, and whether the dem/dsm model on this ground can be by flood with it
Flood and have important contact.In addition the other waters information in this section of region, is also that the probability being inundated with flood with this section of road is had
Close.
In addition to disaster causes real roads damage, folded covering is caused by high-lager building and hypsography, and it is led
If being associated with dem/dsm information；The fracture that road causes through waters (bridge) is related to waters information.
In addition the actual attribute of fracture zone is in addition to be associated with this six evidences, also with this region in sar shadow
The observation showing on picture is direct correlation.The observation that the road that different situations causes breaks on sar image is
Otherwise varied, observation to be represented with gray scale and texture under normal circumstances, and accompanying drawing 4 is the road path loss that the present invention is taken
The Bayesian network model ruined.
The main object of the present invention is to combine actual observed value judgement fracture under the auxiliary of various evidences to calculate
Area belongs to various situation probabilities, so not needing to calculate the joint probability density between all variables.Except outline of house
Vector data, waters contour vector data, the present invention is it is also possible to consider other aucillary documents.Expose in network model shown in accompanying drawing 4
6 priori evident information variable outline of house data (i.e. outline of house vector data after step 2 registration) of going out, sliding
Slope hidden danger point data, calamity kind and its intensity, dsm data, road vectors data, waters information (i.e. waters after step 2 registration
Contour vector data) use a respectively, b, c, d, e, f replacing, 2 observation doubtful road damage area gray scales, doubtful roads
Damage area texture uses g respectively, and representing, hiding doubtful road damage area actual attribute is represented h with x, 6 priori evident information
Variable a, b, c, d, e, f are the conditions of doubtful road damage area actual attribute x, and doubtful road damage area actual attribute x is 2 sights
The condition of measured value g, h.So the present invention needs the conditional probability asked for is p (x/a, b, c, d, e, f, g, h), its computing formula
For:
Wherein p (x/a, b, c, d, e, f) represents that this doubtful road damage area belongs to certain situation under the conditions of various evidences
Prior probability.This conditional probability is determined by expertise, pre-enters as known conditions in the present invention.
By a, this six aucillary documents of b, c, d, e, f can be obtained by the prior probability p (x/ of doubtful road damage area attribute
a,b,c,d,e,f).Calamity kind typically has earthquake, heavy rain etc., then has landslide, house collapse, flood the reason the road damage causing
Respective function can voluntarily be preset Deng, those skilled in the art as the case may be, such as x belongs to landslide (rockslid) to be caused
The Probability p (x=rockslid/a, b, c, d, e, f) of road damage size may be set to (true by variable c with disaster intensity
Fixed) and by d(dem/dsm data) terrain slope that determines is directly proportional, if simultaneously a belong to landslide point so its probability should
This is bigger；X belongs to the Probability p (x=buildingcollaps/ of the road damage that house collapse (buildingcollaps) causes
A, b, c, d, e, f) size may be set to be directly proportional to disaster intensity (being determined by variable c), and by a(outline of house data) certainly
Fixed fracture zone is inversely proportional to the distance in house；X belongs to the Probability p (x=flood/a, b, c, d, e, f) that flood (flood) floods
Size may be set to be directly proportional to disaster intensity (being determined by variable c), and by f(waters information) fracture zone that determines and water body
Distance and by d(dem/dsm data) elevation that determines is inversely proportional to.In addition x belongs to the probability of certain situation with calamity kind (by becoming
Amount c determines) related, it is also different that the natural disaster such as earthquake heavy rain causes the probability of different kinds of roads damage.
Other factors also influence whether that x belongs to the probability of various true damages, such as when according to the judgement of road vectors data e
When learning that this fracture zone is located on intersection, the probability that fracture zone belongs to true damage will reduce, and belong to dry
The probability disturbing information then increases (because intersection is often detected as fracture zone in the present invention the 3rd step)；According to road
When vector data e and waters data f judge to learn that this fracture zone is located on bridge, fracture zone belongs to the general of true damage
Rate will reduce, and the probability belonging to interference information then increases (because when bridge is parallel with radar incident direction, due to not having
Dihedral angle reflection effect, and water body is all very low with the reflex strength of road, may lead to the bridge invisible on sar image)；Root
According to the sensor parameters of dem/dsm and sar image ask for folded cover scope, judge that fracture zone is located at folded when cover scope,
The probability that so fracture zone belongs to true damage is just 0.
P (g/x), p (h/x) represents the relation between the attribute in doubtful road damage area and its observation, Ke Yitong respectively
Cross and solve to calculate this two probability when the distribution that x belongs to observation under different situations respectively, when being embodied as, can will be initial
Probability distribution is set to normal distribution, is then trained with observation and obtains its parameter.Can determine that x belongs to by priori not sympathizing with
The characteristic (comprising the function of unknown parameter) of its gray scale and texture under condition, if there are substantial amounts of observed data, then can be with maximum
The method of likelihood method solves unknown parameter, thus obtaining the probability distribution in the case that x belongs to certain attribute.For example, false
As x is equal to the road damage that landslide causes, then be accomplished by a large amount of observations using landslide causes the region of road damage and enter
Row training, draws the gray scale of the road damage that landslide causes and the probability distribution of texture.
Predetermined p (x/a, b, c, d, e, f) and p (g/x), p (h/x), just can solve based on various evidences and sight
In the case of measured value, the Posterior probability distribution p (x/a, b, c, d, e, f, g, h) of the actual attribute in doubtful road damage area, select
The situation of maximum probability is as final result of determination.
Above content is to further describe it is impossible to assert this with reference to specific embodiment is made for the present invention
Bright it is embodied as being confined to these explanations.For general technical staff of the technical field of the invention, do not taking off
On the premise of present inventive concept, some simple deduction or replace can also be made, all should be considered as belonging to the protection of the present invention
Scope.
Claims (9)
1. a kind of sar image road damage information extracting method based on vector data is it is characterised in that comprise the following steps:
Step 1, according to the scope of the sar image after calamity, obtains the vector data of corresponding region, described vector data includes road
Vector data；
Step 2, after the vector data of step 1 gained corresponding region is projected to the coordinate system of sar image, vector data is joined
Standard is on sar image；
Step 3, extracts the doubtful road damage area of sar image, including following sub-step,
Step 3.1, in the relief area set up according to road vectors data, enters line detection using Road Detection operator, obtains
Road line segment primitive and road width information, find the position of road fracture, obtain according to the testing result of road line segment primitive
First doubtful road damage area；
Step 3.2, step 3.1 gained road width information is combined with road vectors data, obtains shape level-set segmentation
Prior shape constrains, and carries out level set movements using prior shape constraint, obtains the segmentation result of road area and find road
Fracture position, obtain the second doubtful road damage area；
Step 3.3, step 3.1 and 3.2 gained the first doubtful road damage areas and the second doubtful road damage area is merged, obtains
Final doubtful road damage area；
Step 4, sets up the doubtful road damage area that Bayesian network model extracts to step 3 and determines whether, extract
Go out road damage information.
2. according to claim 1 based on vector data sar image road damage information extracting method it is characterised in that:
In step 1, download vector data from openstreetmap server.
3. the sar image road damage information extracting method based on vector data according to claim 1 or claim 2, its feature exists
In: in step 1, described vector data also includes outline of house vector data and waters contour vector data.
4. according to claim 3 based on vector data sar image road damage information extracting method it is characterised in that:
Step 1 includes following sub-step,
Step 1.1,4 angular coordinates of sar image are projected under the coordinate system of vector data, if projection gained longitude and latitude is sat
Mark is respectively (x1,y1), (x2,y2), (x3,y3), (x4,y4)；
Step 1.2, asks for x1,x2,x3,x4Maximum, minima xmax,xminAnd y1,y2,y3,y4Maximum, minima ymax,
ymin；
Step 1.3, by (xmin,ymax), (xmin,ymin), (xmax,ymax), (xmax,ymin) as 4 angle points downloading scope, obtain
Take the vector data of corresponding region.
5. according to claim 4 based on vector data sar image road damage information extracting method it is characterised in that:
In step 3.1, described Road Detection operator is as follows,
The template of Road Detection operator be width be 2w, length be the rectangle of l, the middle section of template is width is w, length
Rectangle for l, the left and right sides region respectively width of template is w/2, length is the rectangle of l；Rung using formwork calculation
Should value gap=min (1-avr2/avr1,1-avr2/avr3), if calculating gap < 0, make gap=0.
6. according to claim 5 based on vector data sar image road damage information extracting method it is characterised in that:
In step 3.1, enter line detection using Road Detection operator, obtain road line segment primitive and road width information, implementation
For if certain width sar image resolution is n rice, making w1=8/n, w2=16/n, w3=24/n, w4=32/n, to road vectors number
According in every section of road vectors set road width w=w successively1,w2,w3,w4, carry out following operation under each value, in this section of road
Road vector both sides is extended, and sets up the relief area that overall width is 4w, and mobile Road Detection operator is detected in relief area,
Record this line segment when response value gap is more than predetermined threshold value；The move mode of mobile Road Detection operator is to swear from this section of road
Amount starting point starts, and caching is divided into several minibuffer areas, the template of each minibuffer section length and mobile Road Detection operator
Length is consistent；In each minibuffer area, from road vectors centrally along perpendicular to road vectors direction toward two side shiftings, width
Often it is separated by w/4 to calculate once；Record in the template accordingly moving Road Detection operator under the result of calculation that each exceedes threshold value
Heart line segment, and calculate the line segment sum detecting in relief area；
The line segment sum relatively detecting in relief area under each value, selects the value detecting most line segments to be road
Developed width.
7. according to claim 6 based on vector data sar image road damage information extracting method it is characterised in that:
In step 3.2, to every section of road vectors in road vectors data, using this section of road vectors as centrage, according to the width of road
Spend the shape constraining information as road for the long strip type region being extended to a closure toward both sides, the length in long strip type region is certain
The length of Duan Daolu, width is the width of road.
8. according to claim 7 based on vector data sar image road damage information extracting method it is characterised in that:
In step 4, described Bayesian network model includes 6 priori evident information variable a, b, c, d, e, f, 2 observations g, h and doubting
Like road damage area actual attribute x, 6 priori evident information variable a, b, c, d, e, f are doubtful road damage area actual attribute x
Condition, doubtful road damage area actual attribute x is 2 observations g, the condition of h；By solving the reality in doubtful road damage area
The Posterior probability distribution of border attribute is as follows, the maximum situation of select probability as final result of determination,
Wherein, p (x/a, b, c, d, e, f) represents that doubtful road damage area belongs to the elder generation of certain situation under the conditions of various evidences
Test probability, p (g/x), p (h/x) represents the relation between the attribute in doubtful road damage area and observation respectively.
9. according to claim 8 based on vector data sar image road damage information extracting method it is characterised in that:
Described 6 priori evident information variable a, b, c, d, e, f are respectively outline of house vector data, landslide point data, calamity kind
With its intensity, dsm data, road vectors data, waters contour vector data；2 observations g, h is respectively doubtful road damage
Area's gray scale, doubtful road damage area texture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310750917.9A CN103714339B (en) | 2013-12-30 | 2013-12-30 | SAR image road damaging information extracting method based on vector data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310750917.9A CN103714339B (en) | 2013-12-30 | 2013-12-30 | SAR image road damaging information extracting method based on vector data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103714339A CN103714339A (en) | 2014-04-09 |
| CN103714339B true CN103714339B (en) | 2017-01-18 |
Family
ID=50407296
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310750917.9A Active CN103714339B (en) | 2013-12-30 | 2013-12-30 | SAR image road damaging information extracting method based on vector data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103714339B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104050473B (en) * | 2014-05-20 | 2019-07-19 | 中国人民解放军理工大学 | A kind of road data extracting method based on rectangular neighborhood analysis |
| CN104091369B (en) * | 2014-07-23 | 2017-02-22 | 武汉大学 | Unmanned aerial vehicle remote-sensing image building three-dimensional damage detection method |
| CN104268879B (en) * | 2014-09-28 | 2017-03-01 | 民政部国家减灾中心 | Building quantity of goods produced Damage assessment method based on remote sensing multispectral image |
| CN104408437B (en) * | 2014-12-08 | 2018-09-21 | 电子科技大学 | A kind of Approach for road detection based on synthetic aperture radar |
| CN106169086B (en) * | 2016-07-21 | 2019-04-09 | 武汉大学 | High-resolution optical image under navigation data auxiliary damages method for extracting roads |
| CN106295699A (en) * | 2016-08-11 | 2017-01-04 | 北京师范大学 | A kind of earthquake Damage assessment method and apparatus based on high-definition remote sensing data |
| CN108229273B (en) * | 2017-02-27 | 2021-03-05 | 北京市商汤科技开发有限公司 | Method and device for training multilayer neural network model and recognizing road characteristics |
| CN108388828A (en) * | 2017-07-13 | 2018-08-10 | 中国科学院遥感与数字地球研究所 | A kind of seashore wetland land cover pattern information extracting method of comprehensive multi- source Remote Sensing Data data |
| CN107886505B (en) * | 2017-11-08 | 2019-06-04 | 电子科技大学 | A kind of synthetic aperture radar airfield detection method based on line primitives polymerization |
| CN108645342B (en) * | 2018-04-25 | 2020-07-07 | 国交空间信息技术（北京）有限公司 | Road width extraction method based on road track and high-resolution image |
| CN108921076B (en) * | 2018-09-21 | 2022-02-18 | 南京信息工程大学 | Pavement crack disease self-adaptive constant false alarm detection method based on image |
| CN109785307B (en) * | 2019-01-09 | 2020-08-07 | 武汉大学 | Unmanned aerial vehicle image road damage assessment method based on vector guidance |
| CN109828999A (en) * | 2019-01-24 | 2019-05-31 | 厦门大学 | Road service system detection recognition method after calamity based on the cross-domain city big data of multi-source |
| CN110176000B (en) * | 2019-06-03 | 2022-04-05 | 斑马网络技术有限公司 | Road quality detection method and device, storage medium and electronic equipment |
| CN111260637B (en) * | 2020-01-19 | 2022-12-13 | 石家庄铁道大学 | Asphalt road transverse fracture detection method based on regional difference |
| CN111582280B (en) * | 2020-05-11 | 2023-10-17 | 吉林省森祥科技有限公司 | Data deep fusion image segmentation method for multispectral rescue robot |
| CN112241806B (en) * | 2020-07-31 | 2021-06-22 | 深圳市综合交通运行指挥中心 | Road damage probability prediction method, device terminal equipment and readable storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831423A (en) * | 2012-07-26 | 2012-12-19 | 武汉大学 | SAR (synthetic aperture radar) image road extracting method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102144173B (en) * | 2008-05-29 | 2014-08-20 | 电视广播有限公司 | Target detection in a sar-imaged sea area |
-
2013
- 2013-12-30 CN CN201310750917.9A patent/CN103714339B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831423A (en) * | 2012-07-26 | 2012-12-19 | 武汉大学 | SAR (synthetic aperture radar) image road extracting method |
Non-Patent Citations (4)
| Title |
|---|
| Road Extraction from High-Resolution SAR Image on Urban Area;C. Han et al.;《2006 IEEE International Symposium on Geoscience and Remote Sensing》;20061231;第1454 - 1457页 * |
| SAR图像道路检测及其仿真;梅珍 等;《计算机仿真》;20100531;第27卷(第5期);第201-203,251页 * |
| 基于道路网矢量数据的遥感影像道路损毁检测;徐阳 等;《测绘通报》;20111231(第4期);第14-16,22页 * |
| 基于高分辨率遥感影像的汶川地震道路损毁评估;陈世荣 等;《遥感学报》;20081130;第12卷(第6期);第949-955页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103714339A (en) | 2014-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103714339B (en) | SAR image road damaging information extracting method based on vector data | |
| Kako et al. | Estimation of plastic marine debris volumes on beaches using unmanned aerial vehicles and image processing based on deep learning | |
| Trier et al. | Automatic detection of mound structures in airborne laser scanning data | |
| Lechner et al. | Remote sensing of small and linear features: Quantifying the effects of patch size and length, grid position and detectability on land cover mapping | |
| CN106127204A (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN110009010A (en) | Wide area optical remote sensing target detection method based on the re-detection of interest region | |
| KR102399089B1 (en) | A system and method for flood damage detection using satellite image information | |
| Mora et al. | Small landslide susceptibility and hazard assessment based on airborne lidar data | |
| Lambers et al. | Towards detection of archaeological objects in high-resolution remotely sensed images: the Silvretta case study | |
| Rogers et al. | VEdge_Detector: automated coastal vegetation edge detection using a convolutional neural network | |
| Weidner et al. | Classifying rock slope materials in photogrammetric point clouds using robust color and geometric features | |
| CN116246169A (en) | SAH-Unet-based high-resolution remote sensing image impervious surface extraction method | |
| Uzar et al. | Automatic building extraction using LiDAR and aerial photographs | |
| Xu et al. | Evaluating urbanization and spatial-temporal pattern using the DMSP/OLS nighttime light data: a case study in Zhejiang Province | |
| CN116205522B (en) | Landslide vulnerability evaluation method and system based on multidimensional CNN coupling | |
| Weidner et al. | The influence of training data variability on a supervised machine learning classifier for Structure from Motion (SfM) point clouds of rock slopes | |
| Kouchi et al. | Damage detection based on object-based segmentation and classification from high-resolution satellite images for the 2003 Boumerdes, Algeria earthquake | |
| CN113128559A (en) | Remote sensing image target detection method based on cross-scale feature fusion pyramid network | |
| Lian et al. | End-to-end building change detection model in aerial imagery and digital surface model based on neural networks | |
| Alba-Flores | Evaluation of the use of high-resolution satellite imagery in transportation applications | |
| CN106339423A (en) | Method and device for dynamically updating sugarcane planting information | |
| Mas | An artificial neural networks approach to map land use/cover using Landsat imagery and ancillary data | |
| White et al. | Automated Land Cover Change Detection Through Rapid Uas Updates of Digital Surface Models | |
| Samodra et al. | Benchmarking landslide inventory data handling strategies for landslide susceptibility modeling based on different random forest machine learning workflows | |
| Islam | Bank erosion and movement of river channel: A study of Padma and Jamuna Rivers in Bangladesh using remote sensing and GIS |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |